feat(cli): rerun captured task bundles

[christso]

Summary

• Add agentv runs rerun <run-dir> to execute captured task bundles from a run index.jsonl using native bundle files (task/EVAL.yaml, task/targets.yaml, task/files/, task/graders/).
• Support selection by --test-id and --source-target, target replacement via --targets / --target, dotenv loading with --env-file, explicit --output, and safe default output placement outside the captured source bundle.
• Preserve rerun traceability by carrying source run/test metadata into normal AgentV result metadata.

Bead

• Bead: av-wy0.4

Base / Dependencies

• Rebased onto origin/main at 35263cd707a9a89c95728ae86beb7271b76f2358.
• The requested prerequisite PRs are merged into this base:
  • docs(agents): remove local orchestration guidance #1334 -> 8e6dd1e96f0ac23c5a413768c82403c0535bf905
  • fix(results): harden remote sync dogfood #1332 -> 083e08c39492b030879a311801d17f6631e909f1
  • feat(trace): add normalized trajectory contract #1331 -> 35263cd707a9a89c95728ae86beb7271b76f2358
• PR feat(results): materialize per-test task bundles #1330 (feat/av-wy0.3-task-bundles) is also merged into main and remains the native bundle foundation for this PR.
• PR feat(cli): simplify eval output surface #1336 (feat(cli): simplify eval output surface) changes eval output flag validation, but runs rerun passes rawOptions.output as a run directory and does not use removed eval flags (--out, --artifacts, --output-format). feat(cli): rerun captured task bundles #1335 can proceed independently; if feat(cli): simplify eval output surface #1336 merges first, rerun behavior remains aligned with --output <run-dir>.

Verification

Post-final-rebase local checks:

• bun run --filter @agentv/core build
• bun test apps/cli/test/commands/runs/rerun.test.ts apps/cli/test/eval.integration.test.ts (17 pass, 0 fail)
• bun --filter agentv typecheck
• bunx biome check apps/cli/src/index.ts apps/cli/src/commands/runs/index.ts apps/cli/src/commands/runs/rerun.ts apps/cli/src/commands/eval/run-eval.ts apps/cli/src/commands/results/manifest.ts apps/cli/test/commands/runs/rerun.test.ts apps/cli/test/fixtures/mock-run-evaluation.ts
• GitHub CI on fa83b2cc16f2af686fb8fd18be2d3e9338ca7d43: Build, Typecheck, Lint, Test, Check Links, Validate Marketplace, Validate Evals, and Cloudflare Pages all succeeded.

Earlier full-branch verification before the unblock rebase:

• bun run verify
• bun run validate:examples (58 valid, 0 invalid)

Manual Red/Green UAT

• Red on pre-feature main: bun --no-env-file /tmp/agentv-av-wy0-4-red/apps/cli/src/cli.ts runs rerun /tmp/nonexistent exited with Not a valid subcommand name for runs.
• Green on this branch: agentv runs rerun against a temp captured bundle wrote a separate output run, preserved test_id: case-alpha, emitted metadata.rerun_source, and regenerated Alpha answer instead of replaying the captured Captured old answer.

Coordination Notes

• Code/test files were reserved in Agent Mail before editing and during the unblock rebase.
• This PR should be reviewed but not merged by the worker.

[cloudflare-workers-and-pages]

Deploying agentv with    Cloudflare Pages

Latest commit:	fa83b2c
Status:	✅  Deploy successful!
Preview URL:	https://28e27d30.agentv.pages.dev
Branch Preview URL:	https://feat-av-wy0-4-rerun-bundles.agentv.pages.dev

View logs