Two small fixes: OTA update-check guard (undefined definition) + permitJoin countdown reset on close

[krobipd]

Two small, independent fixes, each triggered by a real event on a production ember coordinator (EmberZNet 8.0.2). Both are one-line changes; eslint passes. No test scenarios added (small changes) — the production logs below are the "seen it happen".

They surfaced on the same instance but are unrelated (different subsystems) — happy to split into two PRs if you'd rather.

---

1. fix(pairing) — permitJoin countdown not reset on close

Symptom (same session, ~16 min apart):

16:08:06 info: Opening network.
16:08:06 info: Opened zigbee Network for 60 seconds
16:09:06 info: Closed Zigbee network with 1 second remaining.
16:25:06 info: Extending open network.        <-- network had closed 16 min earlier
16:25:07 info: Opened zigbee Network for 60 seconds

Opening the network a second time logs "Extending open network" although the previous window had already timed out and closed — the adapter still considers a pairing window active long after the radio closed.

Cause: in handlePermitJoinChanged, the timeout-close branch clears the countdown interval but never resets _permitJoinTime (it is left at ~1 — the value shown in "1 second remaining"). permitJoin() then treats _permitJoinTime > 0 as an already-open window and logs "Extending …" instead of "Opening …". The manual stop path (permitJoin(0)) already resets it; only the timeout path did not.

Fix: reset _permitJoinTime = 0 when the network closes.

---

2. fix(ota) — update-availability check throws for devices without a definition

Symptom (repeats every OTA cycle for un-interviewed devices):

warn: Failed to check if update available for 'unnamed (0xb4e3f9fffe56ebc8)' Cannot read properties of undefined (reading 'ota')
warn: Failed to check if update available for 'unnamed (0x0c4314fffe0ed00e)' Cannot read properties of undefined (reading 'ota')

A device that never completed interview has no resolved definition (entity.mapped is undefined). lib/ota.js accessed entity.mapped.ota directly, so the check throws and is logged as a misleading "Failed to check if update available" warning.

Cause / consistency: that line was the only entity.mapped.* access in the function not using optional chaining — the four adjacent accesses already use entity?.mapped?.model.

Fix: entity?.mapped?.ota. Such devices now take the existing not_supported branch (which already logs a clean debug line with the id), and the warn is reserved for genuine check failures (device didn't respond).

---

Found on ember; neither change is stack-specific (both are adapter-side JS).