[COVAL-7144] Reconcile the CLI with the Oct 5 API catalog and release 0.9.0

[callumreid]

Reconciles the CLI with the 2026-10-05 public API catalog (183 operations, 437 request fields) and releases 0.9.0.

⚠️ Merge before Mon 2026-10-12 10:00 UTC (03:00 PT). The weekly parity workflow rebuilds chore/weekly-api-parity from main and force-pushes whenever the report differs. That would drop these hand commits and reset this title and body; it already erased the 9/21 hand commit on 9/28.

Changes

• New request fields
  • Metric: Agentic judge judge_mode / enabled_tools; SQL and custom-trace aggregation_method / unit / value_source / span_name / metric_attribute; audio presets and thresholds; metric_metadata.
  • Persona advanced settings: silent_mode, multi_phone_config, initialization_parameters, custom_persona_data, voice, custom_voice_id.
  • Review annotation: structured ground truth.
  • Review project: label configuration, and project_type on update.
  • Run template: test_case_ids.
• Explicit null clears on PATCH. These forward null instead of dropping it:
  • persona advanced settings;
  • run-template test_case_ids (the API rejects [], so null is the only way to remove a test-case subset);
  • metric aggregation_method / unit;
  • review-annotation annotations / ground_truth_json / ground_truth_set_value.
• Removed: review-annotations create --priority. The create contract no longer accepts priority and the server ignored it, so the flag did nothing. Scripts that pass it now get a clap error. update --priority is unchanged.
• Removed: the unused persona_metrics launch field. The CLI always sent None, so the wire format doesn't change.
• Coverage manifest: removes 15 exceptions the spec now covers, including the run-template singular-ID drift from COVAL-5825. Records complete-conversation and sofia/delegation-token as reviewed gaps.
• Version: 0.8.3 → 0.9.0 (minor, because a flag is removed). Merging triggers Release on version bump.

Validation

• cargo fmt --check and cargo clippy --all-targets -- -D warnings: clean.
• cargo test -- --test-threads=2: 9 + 194 passed.
  • In a fully parallel run on a loaded host, test_update_check_notifies_when_outdated can trip its 500 ms budget. That flake predates this PR.
• New explicit-null tests fail on the previous models (positive control) and pass now.
• Audit script tests: OK. Live audit: PASS, 119/183 operations, 437/437 request fields. The regenerated api-coverage-report.md is byte-identical to the committed one. release_version.py reports v0.9.0.
• Live api.coval.dev/v1/openapi is byte-identical to docs origin/main for all 23 specs. Every added field's name and optionality matches the spec.

Follow-ups

• Docs still list review-annotations create --priority (cli/human-review.mdx, llms-full.txt). Remove it when 0.9.0 ships.
• The workflow should stop overwriting a rolling branch that has non-bot commits.

Part of COVAL-2079 (the CLI/API lockstep umbrella stays open).

🤖 Generated with Claude Code

[kdmelon]

APPROVE — first review.

Severity counts: P0 0, P1 0, P2 0, P3 0.

Validation: inspected the complete current-head diff and request serialization paths; all current-head CI, CodeQL, and cross-platform builds are passing. No existing review threads.

Verification limitations: this is a client contract-parity change; review used the PR's mock-server serialization coverage and current-head CI rather than a live authenticated API round trip.