Add GCP Cloud Run OpenTelemetry sample

[eamsden]

Summary

• add a project-neutral Google Cloud Run worker-pool sample using temporalio.contrib.gcp.OpenTelemetryPlugin
• run the Google-Built OpenTelemetry Collector as a startup-probed sidecar, with Secret Manager-backed Temporal credentials and collector configuration
• use the plugin's coordinated 60-second metric default without an override, export cumulative metrics without a collector batch processor, and retain a dedicated five-second trace batch
• demonstrate graceful worker shutdown and bounded Python trace flushing
• pin SDK PR Add GCP OpenTelemetry plugin sdk-python#1658 at 6eeb6f1f38888fce5eab7fdf58c439bcd8d2e163 until the plugin is released

Validation

• poe format
• poe lint (Ruff and mypy; 546 source files)
• mise exec uv@latest -- uv run pytest tests/gcp_open_telemetry -q (6 passed)
• full repository test suite (67 passed, 4 skipped)
• mise exec uv@latest -- uv lock --check
• mise exec uv@latest -- uv lock --project gcp_open_telemetry --check
• git diff --check
• isolated Cloud Run worker-pool and Temporal Cloud E2E verified the unoverridden 60-second default, workflow and activity execution, metrics and trace export, GCP resource attribution, and clean trace/collector shutdown without duplicate-series or exporter errors; the pool was returned to zero instances

Python currently exposes no public Temporal Core metric force-flush API. The sample therefore documents periodic metric export and a clean shutdown window without claiming a separate shutdown metric point. The unbatched metrics pipeline remains safe if the runtime emits a shutdown-time export.

This remains a draft until the SDK plugin is released and the temporary source pins can be replaced with the released dependency version.