fix(slack): require per-date evidence before crediting raw agentic-traffic projection

[akshaymagapu]

Summary

The `check agentic traffic db status` Slack command was permissively crediting any raw `wrpc_import_agentic_traffic` projection_audit row in the lookup window as "raw OK for dateStr", regardless of which date the row actually covered. Raw rows carry no per-date metadata, so a raw run for an earlier date inside the window was indistinguishable from a raw run for `dateStr` itself.

This produced false "raw OK, daily missing" diagnostics. Concrete example from the 2026-05-25 report:

• `sskm.de` and `wknd.enablementadobe.com` were reported as `missing: daily` only, but the DB shows neither raw nor daily had any rows for 2026-05-25 — both were stale by 1-2 days. The raw side was credited because earlier-date raw runs happened to fall in the audit lookup window.

Fix

Raw is now credited for `dateStr` only via one of two date-specific signals:

1. Daily refresh metadata proof: a daily refresh audit row whose `dailyRefreshDates` includes `dateStr`. Daily refresh is a `postSuccessMessage` from raw (see projector `agenticTrafficAnalyticsConfig`), so a daily-for-X row proves raw succeeded for X.
2. Fallback for the legitimate "raw OK, daily not yet run" case: a raw audit row whose `projected_at` lands on or after `dateStr + 1 day` UTC (the raw projector runs the day after the traffic day) with `output_count > 0`.

Also adds `output_count` to the projection_audit select.

Test plan

• `npx mocha test/support/slack/commands/check-agentic-traffic-db-status.test.js` — 14/14 pass
• Updated existing test that asserted the old (buggy) behavior
• New regression test: raw + daily both flagged missing when neither has per-date evidence
• New regression test: raw correctly counted via post-day `projected_at` + positive `output_count` fallback
• New regression test: raw NOT counted when `projected_at < dateStr+1d` even with positive `output_count`
• `npx eslint` clean on both files

🤖 Generated with Claude Code