fix(daemon): do not report trace ingest enqueue during shutdown as a worker crash

[posthog]

Problem

• A normal daemon stop or restart can send a false "crash" to error tracking. The team then triages an issue that is not a crash.
• The trace ingest worker exits on the shutdown signal. A trace connection that is still open can then send one more payload to the closed channel.
• enqueue_trace_payload did not check is_shutting_down(). It logged trace ingest queue send failed: worker may have crashed at error level, and the tracing layer turned that log into an exception.
• Data loss is at most one late trace payload during a stop. This change does not alter that.

Changes

• When the send fails and a shutdown is already in progress, the daemon logs at debug (reason = ingest_worker_stopped_for_shutdown) and returns an error. It does not call request_shutdown again.
• When the channel closes and no shutdown was requested, behavior does not change: error log, "worker may have crashed", and request_shutdown.
• The caller behavior does not change. The trace connection loop still stops on the returned error.

Case	Before	After
Payload after request_shutdown, worker gone	error log + exception, second request_shutdown	debug log, no exception
Worker gone, no shutdown requested	error log + exception, request_shutdown	same

Note

A sequence fault (missing_ingest_seq, duplicate_ingest_seq, reorder_buffer_overflow) calls request_shutdown before the worker exits. Later payloads therefore log at debug. The sequence fault still logs its own error, so error tracking still sees the real fault.

Testing

• Added enqueue_after_shutdown_is_not_reported_as_worker_crash: starts the worker, requests shutdown, waits for the channel to close, then enqueues a payload. It checks the shutdown error and that the queued counter is back to 0.
• Added enqueue_after_unexpected_worker_exit_reports_crash: closes the channel without a shutdown request. It checks the crash error and that the daemon starts to shut down.
• cargo test --locked --lib -- daemon::tests::enqueue daemon::tests::request_shutdown: 4 passed.
• rustfmt --check --edition 2024 src/daemon.rs: clean.
• cargo clippy --locked --all-targets: no new findings. The existing warnings, for example the fetch_update deprecation on current stable, are on main too.
• Skipped: the full task test suite (the sandbox has no task runner). I ran only the daemon unit tests above.

Agent context

• The report named autter-runtime, but the code is in autter-cli (src/daemon.rs).
• I did not add a shutdown check before try_send. While the worker drains, sends must still succeed. The check is only on the failure path.

---

Created with PostHog Desktop from this inbox report.

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}