feat: AlloyDB dual-write for gmail_messages

[dtomic15]

Summary

• Adds optional AlloyDB dual-write alongside existing BigQuery writes
• When ALLOYDB_URL env var is set, the scraper writes to both BQ and AlloyDB
• Adds backfill endpoint ({"action": "backfill"}) to copy existing BQ data to AlloyDB
• All AlloyDB operations are non-fatal — BQ path is completely untouched

Why

The Envision OS MCP tool gmail_search_bigquery was repointed to AlloyDB in Phase 268 (BQ retirement), but the gmail-scraper still only writes to BigQuery. This means emails (including flow@envsn.com) are invisible to Envision OS / Claude users.

Deploy steps (after merge)

1. Update Cloud Run with VPC access + AlloyDB secret:

   gcloud run services update gmail-scraper \
     --region=us-central1 --project=claude-mcp-457317 \
     --network=default --subnet=default \
     --vpc-egress=private-ranges-only \
     --update-secrets=ALLOYDB_URL=envision-alloydb-url:latest

2. Trigger backfill to sync existing BQ data:

   curl -X POST <service-url> \
     -H "Content-Type: application/json" \
     -d '{"action": "backfill"}'

Test plan

• Verify existing BQ writes still work (check Cloud Run logs for BigQuery insert messages)
• Verify AlloyDB dual-write logs appear after VPC + secret configured
• Run backfill for flow@envsn.com: {"action": "backfill", "user_email": "flow@envsn.com"}
• Query via Envision OS: gmail_search_bigquery(user_email="flow@envsn.com")

🤖 Generated with Claude Code