fix(metrics): run metadata backfill once

[svarlamov]

Summary

• persist a completion marker after the legacy metrics metadata backfill reaches the end successfully
• skip the full-table backfill scan on every later daemon startup
• preserve bounded batch processing and retry the migration if any batch fails

Why

The backfill query scans the full metrics table when no legacy rows remain. On a 5.1 GB / 1.48M-row database this held the metrics DB mutex for 8-12 seconds on every daemon restart and caused telemetry flush exceeded its scheduling interval warnings. Current metric inserts already populate the migrated metadata columns, so the historical backfill is one-time work.

Validation

• strict TDD: focused test failed before implementation and passes afterward
• task test TEST_FILTER=backfill_event_metadata
• task lint
• task fmt
• live 5.1 GB database: first fixed startup completed the final scan and persisted the marker; second startup uploaded metrics in ~3.2s with no scheduling warning
• full local suite passes aside from existing environment-specific tests that assume droid is absent and a pre-existing wltrace timing failure on latest main

[svarlamov]

• fix(metrics): run metadata backfill once #2094 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.