Add deterministic CI check-state classifier for reference-server maintenance

[davidahmann]

Reference-server maintenance workflows can misinterpret CI state, causing incorrect escalation or missed regressions.

Current behavior is insufficient because check interpretation is ad hoc and lacks a normalized machine-readable contract.

Why now: cross-repo operations require deterministic CI state categories (failed, pending, no checks, policy-blocked).

Claim-to-codepath map:

• CI triage utilities in repo scripts/package tasks.
• Classification logic and tests in server tooling test suites.

Requested behavior: implement a stable classifier with fixture-backed tests that cover representative PR check scenarios.

Evidence Packet

• Commit under test: a83b145
• Runtime environment:
  • OS: Darwin 25.3.0 arm64
  • Python 3.14.0
  • go version go1.25.7 darwin/arm64
  • Node: v22.19.0
• Minimal repro:
  1. Follow the repo's current local validation/CI-triage path.
  2. Observe the behavior described above for this issue.
• Expected behavior: deterministic, documented behavior matching acceptance criteria below.
• Actual behavior: behavior can be inconsistent for the scenario described above.

Acceptance Criteria

• Implement the smallest code change that addresses the proven pain point.
• Add focused regression test coverage for the changed behavior.
• Ensure backward compatibility expectations are explicit in tests/docs.