Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@nng/CHANGELOG.adoc`:
- Line 21: Remove the personal reminder "(NOTE TO SELF: make a replacement)"
from the changelog entry "* Anything and everything related to options (NOTE TO
SELF: make a replacement)" and either delete that parenthetical entirely or
replace it with a public tracking reference (e.g., "see issue `#NNNN`" or "tracked
by TODO-ISSUE") so the entry is professional; if you prefer creation of an
issue, open one to track the options replacement work and reference its
identifier in the modified entry.

In `@nng/src/lib.rs`:
- Around line 169-171: The public struct Params is annotated with
#[non_exhaustive], which requires Rust 1.40.0 but the crate advertises MSRV
1.36.0; either remove the #[non_exhaustive] attribute from the Params
declaration to retain 1.36 compatibility, or instead update the crate MSRV
metadata (Cargo.toml rust-version / README and CI/tooling that enforce MSRV) to
1.40.0+ and adjust CI/toolchain tests accordingly; reference the Params struct
and the #[non_exhaustive] attribute when making the change.

---

Outside diff comments:
In `@nng/src/device.rs`:
- Around line 81-94: In the reflector function (in nng/src/device.rs) fix the
typo in the unreachable! message: replace "errror" with "error" so the
unreachable!("nng_device returned with no error") uses the correct spelling;
locate the unreachable! call near the nng_device invocation and update that
string accordingly.
- Around line 44-55: Fix the typo in the unreachable! message in the nng_device
completion path: change the string "nng_device returned with no errror" to
"nng_device returned with no error" in the block that calls nng_sys::nng_device
(the code that sets rv and drops s1/s2), ensuring the unreachable! macro uses
the correct spelling; refer to variables rv, s1, s2 and the nng_device call to
locate the statement.

In `@nng/src/lib.rs`:
- Around line 125-160: The PR removed the public nng::options module and its
traits/enums which breaks semver; either reintroduce a compatibility shim or
perform a major version bump—fix by restoring a public options module that
re-exports or wraps the new API with deprecated shims (e.g., recreate pub mod
options that re-exports the old trait/enum names and implements them in terms of
the new Params/additive API), mark those types/functions #[deprecated] with
guidance to migrate, and ensure symbols referenced in tests/clients (the removed
traits/enums and the module name options) are available so cargo semver-checks
passes until a coordinated major release is done.
- Around line 125-259: The thread-count fields on Params (num_task_threads,
max_task_threads, num_expire_threads, max_expire_threads, num_poller_threads,
max_poller_threads, num_resolver_threads) use NonZeroU16 which allows values >
i16::MAX to silently wrap when converted; change their types to NonZeroI16,
update the From<Params> for nng_sys::nng_init_params implementation to use let
convert = |n: NonZeroI16| n.get(); and remove the u16->i16 cast so values map
directly into nng_init_params without wraparound.

---

Nitpick comments:
In `@nng/src/ctx.rs`:
- Around line 158-164: The code in ctx.rs uses a fully qualified enum variant
`nng_sys::nng_err::NNG_ECLOSED` while `nng_err` is already imported elsewhere;
update the assertion in the context close path (the block that calls
`nng_ctx_close(self.ctx)` inside the Drop/close logic) to use the same
qualification style as other files by replacing `nng_sys::nng_err::NNG_ECLOSED`
with `nng_err::NNG_ECLOSED` so it matches uses in `listener.rs` and maintains
consistent imports and style across the crate.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@nng/src/lib.rs`:
- Around line 246-258: The conversion closure currently casts NonZeroU16 to i16
which overflows for values > 32767; update the From<Params> for
nng_sys::nng_init_params implementation to clamp/saturate each NonZeroU16 to
i16::MAX instead of direct casting (e.g., replace convert = |n: NonZeroU16|
n.get() as i16 with a clamp that takes min(n.get(), i16::MAX as u16) and then
casts to i16), leaving map_or(0, ...) semantics intact so out-of-range inputs
become i16::MAX; alternatively, if you prefer a stricter API change, change
Params to use NonZeroI16-backed types and map them directly to the
nng_init_params fields.

In `@nng/src/socket.rs`:
- Around line 77-86: Reformat the multi-line block around the rv2res! expansion
and the unsafe nng_fini() call to match rustfmt expectations: run cargo fmt and
ensure the block that constructs Socket (the unsafe nng_init(ptr::null()) call
inside the rv2res! expansion and the Socket { inner: Arc::new(Inner { handle:
socket, pipe_notify: RwLock::new(None) }) } structure) is formatted as rustfmt
dictates, and similarly format the standalone unsafe { nng_sys::nng_fini(); }
statement; use the rv2res! macro, nng_init, nng_fini, Socket, Inner, and
pipe_notify identifiers to locate the code to reformat.
- Around line 610-614: The Drop impl for the socket calls unsafe
nng_sys::nng_fini() before self.close(), which can free global NNG resources
while the socket still needs them; modify the Drop for the Socket (drop method)
to first call self.close() to properly release socket resources and only after
successful close call unsafe { nng_sys::nng_fini(); }, preserving the SAFETY
comment and any required unsafe block semantics so the global finalizer runs
after per-socket cleanup.

---

Outside diff comments:
In `@nng/src/device.rs`:
- Line 53: There are two typoed panic messages using unreachable! with
"nng_device returned with no errror"; locate the unreachable! invocations in
nng/src/device.rs (the calls associated with the nng_device handling) and change
the string to "nng_device returned with no error" in both occurrences so both
panic messages are corrected and identical.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@nng/src/lib.rs`:
- Around line 172-203: The intra-doc links incorrectly reference `Init` (which
doesn't exist) in the comments for `num_task_threads`, `num_expire_threads`, and
`num_poller_threads`; update those links to point to the `Params` type instead
(change `Init::max_task_threads` → `Params::max_task_threads`,
`Init::max_expire_threads` → `Params::max_expire_threads`, and
`Init::max_poller_threads` → `Params::max_poller_threads`) so the documentation
links resolve to the correct symbols associated with the `Params` struct/fields.

In `@nng/src/socket.rs`:
- Around line 694-697: The call to nng_sys::nng_socket_raw currently ignores its
return code and only inspects the mutable raw boolean, which masks failures;
capture the c_int return value from nng_socket_raw(socket.inner.handle, &mut
raw) and check it (e.g., ensure it is 0 / NNG_OK), returning or converting the
real error instead of proceeding to produce a CookedSocketError when the socket
handle is invalid/closed; update the code around socket.inner.handle, the raw
variable, and the CookedSocketError path to surface or assert on the actual
nng_socket_raw failure.
- Around line 75-86: Move the NNG initialization before any socket open calls
and centralize that logic into a shared helper (e.g., ensure_nng_initialized or
init_nng_once) used by both Socket::new and RawSocket::new so nng_init is always
called exactly once before any nng_* operations; update Inner::drop to only call
nng_fini when the shared init path actually initialized NNG (or rely on an
atomic refcount/Once to balance fini), and modify the RawSocket::new path to
invoke the shared initializer rather than skipping it. Also handle and propagate
the return value of nng_socket_raw (check its nng return code, e.g., map
non-zero to an Err) instead of assuming it always succeeds so errors like
NNG_ECLOSED are surfaced.

---

Duplicate comments:
In `@nng/src/lib.rs`:
- Around line 169-171: The #[non_exhaustive] attribute on the public Params
struct raises the effective MSRV above the documented 1.36; either remove the
#[non_exhaustive] annotation from the Params struct declaration in lib.rs or
update the crate MSRV metadata and tests (e.g., bump documented Rust version and
rust-version/Cargo.toml and CI toolchains to >=1.40.0) so the public API is
consistent with using #[non_exhaustive].
- Around line 246-258: The conversion currently casts NonZeroU16 values with
`let convert = |n: NonZeroU16| n.get() as i16;` which will wrap values >
i16::MAX into negatives; update the conversion used in the From<Params> for
nng_sys::nng_init_params implementation to clamp to i16::MAX before casting
(e.g. bound the u16 with .min(i16::MAX as u16) and then cast to i16) so fields
like num_task_threads, max_task_threads, num_expire_threads, max_expire_threads,
num_poller_threads, max_poller_threads, and num_resolver_threads never become
negative across the FFI boundary.

In `@nng/src/socket.rs`:
- Around line 610-614: The nng library finalization (nng_fini) is currently
called before closing the socket, which is unsafe because self.close() (which
calls nng_socket_close) must not be invoked after nng_fini; swap the calls so
that self.close() is invoked first and then the unsafe { nng_sys::nng_fini(); }
is executed, preserving the existing SAFETY comment pairing with socket
initialization and ensuring no NNG calls occur after finalization.