results(summary): rename cases to tests, trim run metadata - #1697

Summary

• Renames run-root summary.json's cases[] to tests[] and aligns counts to test/sample terminology: total_cases/passed_cases/failed_cases → total_tests/passed_tests/failed_tests, total_instances/errored_instances → total_samples/errored_samples.
• Drops the metadata.run_id duplicate of the summary root's run_id field. summaryRunId() (used when rebuilding cross-run indexes) reads the root field first and falls back to the old metadata.run_id location so historical bundles written before this change still resolve correctly.
• Renamed internal variables in buildRunSummaryArtifact (casesByKey→testsByKey, instances→samples, etc.) to match the new terminology.

Implements av-cpl5.2 (child of source-of-truth tracker av-cpl5). This is the writer side only — RunSummaryArtifact, buildRunSummaryArtifact, writeInitialRunSummaryArtifact, and aggregateRunDir's summary-writing path in packages/core/src/evaluation/run-artifacts.ts. Dashboard/CLI reader compatibility for historical bundles (old cases/total_cases names) is scoped to the parallel bead av-cpl5.1; no reader code outside run-artifacts.ts referenced these fields (verified via repo-wide search), so nothing else needed updating in this PR. .internal/index.jsonl and IndexArtifactEntry (av-cpl5.3's territory) were left untouched.

Known follow-up (out of scope here): metadata.eval_file is written as whatever absolute/relative path the CLI passes into buildRunSummaryArtifact/writeArtifactsFromResults — in a live dogfood run it came through as an absolute host path, while the sibling metadata.runtime_source.eval_files is already repo-relative (normalized upstream in run-eval.ts). Fixing eval_file would need cwd/repoRoot plumbing through multiple caller files outside this bead's declared scope; flagging for a follow-up bead rather than expanding this PR's surface.

Test plan

• bun test apps/cli/test/commands/eval/artifact-writer.test.ts — 75 pass, including new/updated assertions for tests[], counts.total_tests/passed_tests/failed_tests/total_samples/errored_samples, and top-level run_id.
• Regression: bun test apps/cli/test/commands/eval/aggregate.test.ts packages/core/test/evaluation/target-execution-artifacts.test.ts packages/core/test/evaluation/orchestrator.test.ts — 115 pass.
• Regression: bun test apps/cli/test/commands/results/report.test.ts apps/cli/test/commands/results/export.test.ts — 39 pass.
• bun --filter @agentv/core typecheck — clean.
• bunx biome check on touched files — clean.
• Live dogfood: bun apps/cli/src/cli.ts eval run examples/features/rubric/evals/operators.eval.yaml --target llm --workers 1 --threshold 0.8 against a live Azure LLM target with a live llm-rubric grader (threshold 0.8, not a 0-threshold smoke check) — passed 100%, and the resulting .agentv/results/<run_id>/summary.json was inspected end-to-end to confirm the new tests[]/counts.total_tests/etc. shape and the removed metadata.run_id duplicate.

🤖 Generated with Claude Code