fix(hooks): give SessionEnd a timeout so shutdown does not cancel it

[overkamperic]

Relates to #13.

Problem

On native Windows the SessionEnd hook is cancelled on every exit (failed: Hook cancelled), and since the final turn is held back until SessionEnd, that turn is never exported. v1.2.0's lazy-import fast path does not avoid it here: starting the hook through uv run --script alone takes about 1.3 s, which together with the processing time exceeds the shared 1.5 s SessionEnd window.

The hooks reference documents the fix: SessionEnd hooks share a 1.5-second budget, and if a hook sets a longer timeout, Claude Code raises the budget to match (up to 60 s). This PR sets "timeout": 10 on the SessionEnd hook only. Stop is unchanged.

Evidence

Environment: Windows 11, Claude Code 2.1.285, uv 0.12.15 (d35f1f270 2026-09-15 x86_64-pc-windows-msvc), plugin 1.2.0, self-hosted Langfuse. Each run was a fresh one-turn claude -p session.

hooks.json	"Hook cancelled" on exit	Final turn exported (hook log)
stock 1.2.0	3 of 3 runs	no (only the Stop run's Processed 0 turns)
with "timeout": 10	0 of 2 runs	yes (Processed 1 turns about 4 s after exit)

The same before/after held on 1.1.0: cancelled without the timeout, clean with it. User settings also had another SessionEnd hook with timeout: 10, and the plugin's hook was still cancelled until the plugin's own entry set one.

Checks

• hooks/hooks.json parses; only the SessionEnd entry changes.
• CI smoke check: echo '{}' | uv run --script hooks/langfuse_hook.py exits 0.
• No Python changes, so the pytest suite was not re-run.

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}