feat(visibility): surface per-collector errors to Flow Doctor

[cipher813]

Summary

• Wires alpha_engine_lib.collector_results.report_collector_errors() into weekly_collector._finalize so per-collector error-status entries emit one logger.error each, with the collector name + original error message.
• Bumps alpha-engine-lib pin v0.5.1 → v0.6.2 (helper added in feat(collectors): report_collector_errors helper for Flow Doctor visibility (v0.6.2) alpha-engine-lib#34).
• Adds wiring test test_collector_error_visibility.py to pin the call site.

Why

The 2026-05-09 Saturday SF DataPhase1 PARTIAL run produced no per-failure Flow Doctor alert despite arcticdb failing on a backfill regression preflight. Tracing showed the only logger.error() that fires on partial status is main()'s end-of-run generic summary:

"Weekly collection finished with non-ok status=partial — Per-collector statuses: {arcticdb: error, ...}"

Two problems:

1. Dedup collision. Same string for every partial run regardless of which collector failed → Flow Doctor's per-message dedup treats all partial runs as one alert. Cooldown window suppresses different-cause partial runs.
2. No diagnose context. The actual error ("Backfill regression preflight failed: 38 symbols would regress...") lives only in the result dict. Flow Doctor's LLM diagnosis + GitHub-issue body has nothing specific to work with.

Fix placement

Call inside _finalize immediately after the status block:

• Fires for every code path that finalizes (phase 1, phase 2, daily, morning enrich) without per-call wiring.
• Runs before manifest write + postflight, so per-collector errors are visible to Flow Doctor even if postflight raises.
• One line of orchestrator wiring, helper does the walking.

Dependency

⚠️ Blocked on cipher813/alpha-engine-lib#34 — that PR adds the collector_results module and tags v0.6.2. Merge order:

1. Lib PR Fetch macro tickers daily via polygon → FRED → yfinance #34 → tag v0.6.2
2. This PR's CI install of @v0.6.2 resolves
3. Merge here

Test plan

• New wiring test (2 cases): per-collector errors emit ERROR records; all-ok runs emit none. Both pass with lib installed locally.
• Full unit suite: 552 passed, 1 skipped.
• Sat 2026-05-16 SF cron will be the first production exercise. If a collector errors, expect a per-collector Flow Doctor alert in addition to the existing PARTIAL email.

Follow-up

• Opportunistic adoption in alpha-engine-research / alpha-engine-predictor where the same status-dict pattern exists in stage outputs / per-team collectors.

🤖 Generated with Claude Code