test: cover ai-evals API drift and CCM anomaly v2 wiring

[cursor]

Description

Adds regression tests for two recently merged production changes that shipped without dispatch/schema contract coverage.

Risky behavior now covered

AI Evals (#529) — prevents reintroduction of control-plane API drift:

• eval_metric create requires and forwards dimension (422 without it)
• eval_annotation create/update expose and preserve thumbs_up (including false)
• online_eval evaluate uses metric_set_id + judge_llm_connector_ref instead of deprecated metric_ids

CCM anomalies (#533) — guards v2 API migration:

• anomalyListExtract projection shape
• cost_anomaly v2 list body builder: defaults, explicit time range vs time_filter, zero thresholds (!= null), ordering/search
• cost_anomaly_drilldown path routing (details vs time-series vs sub-item list) and skipCompact round-trip

Test files added/updated

• tests/registry/ai-evals.test.ts — control-plane drift + online_eval dispatch tests
• tests/registry/ccm-anomaly.test.ts — new v2 anomaly dispatch and drill-down tests
• tests/registry/ccm-extractors.test.ts — anomalyListExtract unit tests

Why these tests materially reduce regression risk

Both fixes changed request body shapes and API paths that agents rely on at runtime. Schema-only checks would miss dispatch bugs (e.g. dropping thumbs_up: false, reverting to metric_ids, or using truthy checks that drop min_amount: 0). These tests assert the exact HTTP path/body the registry emits.

Type of Change

• Bug fix
• New feature
• Refactor
• Documentation
• Other — test coverage

Checklist

• pnpm test passes (67 tests in touched files)
• No production code changes

  

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}