Let metrics degrade gracefully when no deploy workflow is configured

[coygeek]

Summary

deploybot metrics currently fails hard when the configured deployment workflow does not resolve to one active workflow. That makes metrics unusable for repositories that want merge-queue timing but do not have a deployment workflow yet.

For a first pilot or library-only repo, queue metrics such as request-to-queue and queue-to-merge are still useful even when merge-to-live cannot be computed.

Evidence

Running metrics against this repository with the example policy fails because the example policy configures a Deploy workflow but the repository only has CI active:

error: configured workflow 'Deploy' did not resolve to one active workflow

The command attempted:

deploybot --repository Forward-Future/DeployBot \
  --config .mergequeue.example.toml \
  metrics --limit 10 --json

The active workflow list contains only:

[
  {
    "name": "CI",
    "path": ".github/workflows/ci.yml",
    "state": "active"
  }
]

Impact

Metrics become all-or-nothing:

• A repo without deployment cannot view queue throughput.
• A repo with a renamed deploy workflow gets no partial timing report.
• During onboarding, the user cannot use metrics to prove merge-queue improvements until deployment is fully wired.

This is especially awkward because DeployBot can be valuable before production deployment is configured: it can still coordinate ready PRs and report queue behavior.

Proposed behavior

If deployment workflows cannot be resolved, deploybot metrics should still return partial samples:

• request_to_queue_seconds
• queue_to_merge_seconds
• request_to_live_seconds = null
• merge_to_live_seconds = null
• a warning field explaining that deploy workflow timing was unavailable

For example:

{
  "sample_count": 10,
  "warnings": [
    "configured deploy workflow 'Deploy' did not resolve to one active workflow; live timing omitted"
  ],
  "queue_to_merge_seconds": {
    "count": 8,
    "p50": 42.0,
    "p95": 120.0,
    "max": 120.0
  },
  "merge_to_live_seconds": {
    "count": 0,
    "p50": null,
    "p95": null,
    "max": null
  }
}

The existing hard failure can remain for commands that actually need deployment workflow dispatch or verification. This issue is only about read-only metrics.

Acceptance criteria

• deploybot metrics --json returns queue timing when deploy workflows are missing or unresolved.
• The output includes a machine-readable warning when live/deploy timing is omitted.
• Text mode prints the same warning without a traceback.
• Existing behavior remains fail-closed for merge/follow/react paths that require deployment workflow identity.
• Tests cover missing deploy workflow, renamed deploy workflow, and no merged PR samples.

Verification

• Add/update unit tests for delivery_metrics.
• Run python -m unittest discover -s tests.
• Run ruff check ..