test(cli): simplify churny local coverage

[christso]

Summary

Local test runs now spend less time re-running the same CLI subprocess fixtures for pipeline artifacts. The pipeline command tests still protect the stable contracts users and downstream tooling depend on: manifest fields, code-grader execution, builtin contains/regex/negate grading, grading.json, index.jsonl, and benchmark.json output.

This also moves Dashboard launch-mode and threshold behavior into the docs site so end-user semantics are documented without pinning as many duplicate migration/presentation tests.

What changed

• Collapsed repeated pipeline input, pipeline grade, and pipeline bench subprocess tests so each fixture run asserts all artifacts from that run.
• Kept the full input -> grade -> bench smoke test intact.
• Moved duplicated navigation helper coverage out of the CLI package and kept it beside the Dashboard route helper.
• Reduced duplicated Dashboard threshold/config test cases while preserving canonical dashboard.threshold, legacy fallback, clamping, and save/migration coverage.
• Updated apps/web/src/content/docs/docs/tools/dashboard.mdx with the current Projects-dashboard default and threshold config behavior.

Before/after evidence

Pipeline subprocess test reduction against origin/main:

File	Tests before	Tests after	await execa before	await execa after
apps/cli/test/commands/eval/pipeline/input.test.ts	10	4	10	4
apps/cli/test/commands/eval/pipeline/grade.test.ts	6	3	6	3
apps/cli/test/commands/eval/pipeline/bench.test.ts	5	3	5	3
apps/cli/test/commands/eval/pipeline/pipeline-e2e.test.ts	1	1	3	3

Net for these files: 22 -> 11 test cases, and 24 -> 13 subprocess invocations.

Timing report for next review

I also timed every test file individually after this change. This is not the same as the full package-level pre-push run, but it identifies where local time is concentrated.

• Files timed: 175
• Elapsed wall time for the per-file timing pass: 2.0 min
• Group totals from individual file runs:

Group	Files	Tests	Time
packages/core	112	1787	64.9s
apps/cli	46	543	51.9s
packages/phoenix-adapter	5	21	1.8s
apps/dashboard	6	39	1.1s
packages/eval	5	67	0.9s
plugins	1	7	0.2s

Slowest files after this PR:

Rank	File	Time	Tests	Notes
1	packages/core/test/evaluation/orchestrator.test.ts	9.9s	87	likely biggest remaining core target
2	packages/core/test/evaluation/workspace/pool-manager.test.ts	6.6s	31	real workspace/repo materialization
3	apps/cli/test/eval.integration.test.ts	6.1s	4	full CLI integration subprocesses
4	apps/cli/test/commands/eval/pipeline/input.test.ts	4.7s	4	still subprocess-heavy after reduction
5	apps/cli/test/commands/eval/assert.test.ts	3.9s	4	CLI subprocess assertions
6	apps/cli/test/commands/trend/trend.test.ts	3.9s	10	mix of pure and CLI coverage
7	apps/cli/test/commands/eval/pipeline/grade.test.ts	3.2s	3	still subprocess-heavy after reduction
8	apps/cli/test/commands/eval/pipeline/pipeline-e2e.test.ts	2.9s	1	full pipeline smoke
9	apps/cli/test/commands/eval/pipeline/bench.test.ts	2.9s	3	still subprocess-heavy after reduction
10	apps/cli/test/commands/results/serve.test.ts	2.7s	59	includes git remote/API integration cases

Timing limitation: packages/phoenix-adapter/test/agentv-normalize.test.ts failed only in direct single-file timing because one test resolved an example path outside this worktree. The full repo pre-push test hook passed.

Verification

• bun run build passed. Existing Dashboard bundle-size warning remains.
• bun test apps/cli/test/commands/eval/pipeline/input.test.ts apps/cli/test/commands/eval/pipeline/grade.test.ts apps/cli/test/commands/eval/pipeline/bench.test.ts apps/cli/test/commands/eval/pipeline/pipeline-e2e.test.ts passed: 11 tests.
• bun test apps/cli/test/commands/results/studio-config.test.ts apps/cli/test/commands/results/serve.test.ts apps/dashboard/src/lib/navigation.test.ts passed: 86 tests.
• bun --filter agentv typecheck passed.
• bunx biome check <changed files> passed.
• Pre-push hook passed: Build, Typecheck, Lint, Test, Validate eval YAML files.

---