feat(api): GET /events accepts since and limit query params with X-EXO-Last-Idx header

[5F0jd2vLq54RerYW]

Summary

Extends GET /events to support cursor-based tailing without polling the full /state snapshot.

Changes:

• src/exo/api/main.py: Add optional since (default 0) and limit (default None) query params to stream_events(). Takes a fast-path to read_all() when both are absent — identical to pre-patch behavior.
• src/exo/api/tests/test_stream_events.py: 7 test cases covering full-dump backward compat, since+limit, since-only, since-beyond-count (empty), limit-clamped, negative since rejected (FastAPI ge=0), and chained cursor reads with no overlap.
• src/exo/api/tests/conftest.py: sys.modules stub for exo_rs so API tests run without requiring a compiled Rust extension.

Response header: X-EXO-Last-Idx: <N> is set to the upper exclusive bound of the range consumed, allowing clients to chain reads with no gap and no overlap.

Motivation

Downstream tooling needs to detect state changes efficiently. The previous approach required polling the full /state snapshot (~150KB). With this change, a client can tail events incrementally:

cursor = 0
while True:
    resp = requests.get(f"http://localhost:52415/events?since={cursor}")
    events = resp.json()
    cursor = int(resp.headers["X-EXO-Last-Idx"])
    process(events)
    sleep(1)

Backward Compatibility

GET /events with no query params takes the same read_all() fast-path as before. No existing client behavior changes.

🤖 Generated with Claude Code

[5F0jd2vLq54RerYW]

Note: nix is not available in this dev environment, so I ran ruff format (which is what nix fmt invokes for Python files per the treefmt config in flake.nix) and pushed the formatting commit. The Rust/Svelte/TOML formatters don't apply to this PR (Python-only change).