Add workflow metrics endpoint for workflow-stage and agent-action analytics

[akgohain]

Motivation

• Provide a backend analytics endpoint to surface workflow-stage transitions and agent approval/rejection activity for observability and reporting.
• Compute stable metrics even when event streams are sparse or empty so callers receive predictable schema and safe defaults.

Description

• Add Pydantic models in server_api/workflow/models.py for WorkflowEvent and the WorkflowMetricsResponse schema.
• Implement metrics aggregation in server_api/workflow/service.py with compute_metrics that calculates event counts by type, approvals/rejections counts and rates, stage transition counts, and first/last event timestamps.
• Add a small in-memory event store in server_api/workflow/store.py with get_events, set_events, and clear_events to support seeding and retrieval for the endpoint and tests.
• Expose an APIRouter in server_api/workflow/router.py that provides GET /api/workflows/{workflow_id}/metrics (router is available as server_api.workflow.router and can be mounted into the main app).
• Add endpoint-focused tests in tests/test_workflow_metrics.py that validate safe defaults for empty workflows and correctness against seeded events.

Testing

• Ran PYTHONPATH=. pytest -q tests/test_workflow_metrics.py which succeeded: 2 passed (the test module mounts the router directly for isolation).
• Attempted uv run pytest -q tests/test_workflow_metrics.py which failed in this environment due to package metadata issues for an unrelated editable dependency (pytorch_connectomics), not due to the new workflow tests.

---

Codex Task

[akgohain]

Consolidated into draft PR #165 (feat/codex-wave-integration). Closing this source draft to keep review in one place.