fix(infra): deployment checks and GCP uptime monitor gate on tautological /health 200 #649

Summary

After PR #646 (issue #488), /health is a liveness-only endpoint that answers HTTP 200 by design even when every dependency is down. Two verification layers still treat that 200 as meaningful, so they now pass tautologically on a fully degraded deployment.

Current behaviour

• terraform/modules/deployment-checks/main.tf:60-81: the post-apply check "health" asserts only HTTP 200 on /health; the warmup provisioner at main.tf:35-56 also accepts any non-000/502/503 status. Since /health is 200-always, this gate passes on a fully degraded deploy. Complication: the module also runs against the Lambda platform (terraform/environments/aws/checks.tf:13-17), where /ready does not exist (the route is registered only on the server mux, not the Lambda API router at internal/api/router.go:355-356).
• terraform/modules/monitoring/gcp/main.tf:600-612: the GCP uptime check asserts only HTTP 200 on /health, so it reports 100% availability while every replica is 503 on /ready. (The Azure counterpart content-matches "healthy" in the body and still detects degraded.)

Steps to verify the gap

Point the deployment-checks module or the GCP uptime check at an instance whose database is unreachable: both report success while /ready answers 503.

Expected behaviour

Post-apply deployment checks fail when the deployment is degraded; uptime monitoring can distinguish "process alive" from "serving traffic" where that signal is wanted.

Proposed fix

• Give the deployment-checks module a per-platform health-path variable (/ready for container platforms, /health + body parse for Lambda), or register /ready in the Lambda API router too — noting that Lambda deliberately keeps lazy init, so a Lambda /ready would need to trigger or await initialization rather than just report it.
• For the GCP uptime check, either add a content match on "healthy" (mirroring the Azure monitor) or leave as a pure availability signal by decision; do NOT point uptime monitors at /ready (deploys would flap alerts).

References

• Parent: fix(server/azure): exclude uninitialized replicas from readiness #488, PR fix(server): exclude uninitialized replicas from readiness #646

Severity

Medium — deployment verification and monitoring blind spots, not a live outage vector.