fix(encoding,ci): silence self-healing warmup 503 alert + robust E2E result check

[beveradb]

Summary

Two unrelated production alert-noise fixes, prompted by two false-alarm Discord alerts on 2026-05-24:

• Encoding worker 503 → WARNING. A primary-VM warmup hitting 503 SERVICE_UNAVAILABLE (or capacity exhaustion) is self-healing — the encoding flow's ensure_any_running() falls back to an alternate-zone worker and the job completes (verified live on job a40f3024). It no longer logs at ERROR, so it stays below the error monitor's severity>=ERROR filter and stops paging Discord for a non-incident.
• E2E daily test result check. Both stages now derive pass/fail from Playwright's real exit code instead of grepping the log for "failed". The old check matched the word "failed" inside a benign retry warning (goto attempt 1/3 failed) and reported failure on a run where the job completed and Playwright printed 1 passed.

Changes

• backend/api/routes/encoding_worker.py — catch EncodingWorkerStartError in /warmup and log at WARNING with a clear fallback message; genuine unexpected errors still log at ERROR.
• .github/workflows/e2e-daily.yml — capture ${PIPESTATUS[0]} (the pipe to tee was masking Playwright's exit code as 0) and gate the result on it, for both Stage 1 and Stage 2.
• backend/tests/test_encoding_worker_routes.py — 3 regression tests: 503 start failure → WARNING, capacity error → WARNING, unexpected error → still ERROR.
• docs/TROUBLESHOOTING.md — note that a WARNING-level warmup 503 is benign/self-healing.
• pyproject.toml — bump 0.174.11 → 0.174.12.

Testing

• pytest backend/tests/test_encoding_worker_routes.py backend/tests/test_encoding_worker_manager.py — 46 passed
• YAML validated
• CodeRabbit CLI review — 0 findings

Review

• Local CodeRabbit review completed
• Issues addressed

@coderabbitai ignore

---

🤖 Generated with Claude Code