Summary

This PR reduces App Engine cold-start overhead, simplifies the current single-service deployment path, and fixes regressions uncovered during the startup refactor.

What changed

Startup and runtime

• keeps the app on one service and lazy-loads admin on first /admin access
• keeps GCS imports off the main startup path
• adds startup and per-request timing logs to distinguish cold vs warm behavior
• adds a warmup endpoint and keeps the business-hours warmup strategy
• makes normal API startup independent of SESSION_SECRET_KEY; /admin now returns a clear 503 when that secret is unset

Deploy and local dev

• simplifies Docker Compose back to db + app
• keeps entrypoint.sh parameterized, with migrations enabled by default for local dev
• hardens App Engine CD cleanup so first deploy and single-version cases do not fail
• stops workflows from touching dispatch config

Database and OGC fixes

• adds a follow-up Alembic migration to rebuild ogc_water_elevation_wells for databases that recorded the earlier revision but still have the old materialized view definition
• adds integration coverage for the materialized view schema
• fixes the OGC regression caused by the env-helper refactor by switching query boolean parsing to services.env.to_bool

Code cleanup

• extracts lightweight env parsing into services/env.py
• keeps the low-risk lazy-loading improvements that survived the earlier service-splitting experiment
• adds targeted tests for lazy admin loading and request timing

Validation

Ran:

• pytest tests/test_lazy_admin.py tests/test_request_timing.py tests/test_ogc.py tests/test_admin_views.py tests/integration/test_alembic_migrations.py tests/test_cli_commands.py -q
• result: 55 passed, 3 skipped
• pytest tests/test_ogc.py -q
• result: 12 passed, 2 skipped

Also verified:

• App Engine template and workflow YAML parsing
• targeted compile checks on touched Python files

Notes

• no new required deploy secrets were introduced
• App Engine runtime service account behavior remains aligned with the prior template using CLOUD_SQL_USER
• the main remaining cold-start cost is pygeoapi, not admin