Admin is alerted when the GTFS-RT archiver's per-fetch latency regresses or costs increase substantially

[vevetron]

User story / feature request

As an Admin operating the GTFS-RT archiver,
When a redeploy or dependency change silently doubles per-fetch latency or costs,
I am alerted so I catch the regression in hours instead of days.

Acceptance Criteria

Admin sees that a Cloud Monitoring alert is configured on the GTFS-RT archiver's per-fetch latency — it fires when the median fetch time exceeds ~1.7s (≈1.5× the ~1.1s baseline), sustained ~15 minutes, and notifies some appropriate user group

Notes

• Background: on Aug 5 a one-line redeploy (revert heartbeat.py #5624) rebuilt the archiver image; unpinned google-cloud-* deps drifted and per-fetch latency doubled (~1.1s → ~2.4s, ~$15 → ~$31/day, ~$480/mo). Nothing in prod watched per-fetch latency, so it lurked for days.
• Add a google_monitoring_alert_policy in iac/cal-itp-data-infra/logging/us/, alongside the existing gtfs_low_write_alert; reuse the email_dds_notify channel. Watch run.googleapis.com/request_latencies p50 for service gtfs-rt-archiver.
• Staging's existing SLO alert (99% under 10s) would NOT catch this — 2.4s is under 10s. The guardrail must watch the median, not the tail.
• Optional second condition on run.googleapis.com/container/billable_instance_time to alert directly on the cost driver.
• Cost is negligible (<$1/mo).
• Complementary fixes tracked separately: pin google-cloud-* deps and reuse a module-level storage.Client().