fix: prevent template DAU undercount from session telemetry gaps

[steve8708]

Summary

• Count signed-in activity from legacy and canonical session-status events plus identified app-entry telemetry.
• Repair the persisted BigQuery template DAU query and keep the catalog/seed definitions aligned.
• Add regression coverage for event aliases, fallback activity, and BigQuery dashboard repair.

Root cause

The template DAU panels treated session status as the only signed-in activity signal. From Sep 15 through Sep 19, identified app-entry activity continued while session-status rows fell, so every template was undercounted. The BigQuery panel also did not recognize the canonical session_status alias introduced later. This fix is scoped to the Sep 15+ decline; the known Sep 2-6 zero window is not changed.

Validation

• Analytics focused tests: 111 passed
• Analytics typecheck: source checks clean; local production-config warnings remain for missing deployment secrets/database
• Repository guards: 77 passed

Deployment remains subject to the normal post-merge prebuilt publisher boundary.

[github-actions]

Here's a visual recap of what changed:

Open the full interactive recap

[builder-io-integration]

Builder reviewed your changes — looks good ✅

Review Details

Code Review Summary

This PR fixes an analytics DAU undercount by expanding signed-in activity detection to include fallback event signals.

What Changed

The PR updates template analytics queries across 8 metrics (DAU, WAU, retention, recurring users) to count signed-in activity from:

1. Both session_status and session_status event name variants (handling telemetry aliases)
2. app_entered events with identified users as a fallback when session telemetry gaps occur

Core filter updated from event-based to OR-based logic:

• Before: event_name = 'session status' AND signed_in = 'true'
• After: ((event_name IN ('session status', 'session_status') AND signed_in = 'true') OR (event_name = 'app_entered' AND identified user)) AND user_key IS NOT NULL

What I Verified

Filter Logic — Both PostgreSQL and BigQuery implementations correctly handle OR conditions with proper parenthesization. User key deduplication prevents double-counting across event types. ✓

Backward Compatibility — Legacy SQL constants preserve old behavior. Repair functions safely migrate stale BigQuery queries to new format using two-step replacement (full pattern → partial pattern). ✓

Data Consistency — All 8 affected panels have matching filters in the metric catalog, seed dashboard, and test expectations. Descriptions updated to reflect fallback behavior. ✓

Event Classification — isMarketingWebsiteSessionEvent() now accepts both event name variants. User identification checks properly validate required fields per event type. ✓

Test Coverage — Regression tests validate event aliases, fallback activity, and metric catalog/seed synchronization. All 111 tests passing. ✓

Risk: Standard — This is a scoped business logic fix to analytics query generation with no authentication, payment, or destructive changes involved. Backend-only change with no UI impact.

🧪 Browser testing: Skipped — PR only modifies backend analytics templates and tests, no UI code touched