feat(scenarios): add 3 acceptance scenarios for pr-review and code-review dispatcher

[bborbe]

Summary

Adds three end-to-end acceptance scenarios under `scenarios/` following the dark-factory scenario-writing guide. Each scenario was drafted, audited by `dark-factory:scenario-auditor` twice, and tightened against the audit findings before commit.

#	Title	Validates	Regression risk locked
001	toolchain-preflight	Dispatcher Step 4.0 (PR #36) aborts when ast-grep is absent	coding#34 (silent loop, 30-min activeDeadlineSeconds kill instead of immediate fail)
002	clean-pr-zero-findings	Zero-violation diff in standard mode produces empty findings, no hallucination	LLM tier invents findings to fill an empty mechanical surface
003	scaling-funnel-100-files	100-file synthetic PR completes in ≤30 LLM calls and ≤30 minutes (Phase 10)	coding#27 (30-min activeDeadlineSeconds ceiling)

Four-criteria gate

Each scenario passes the writing-guide's strict bar for adding an E2E scenario:

• Unit/integration cannot reach (real PATH resolution, real slash-command invocation, real Claude CLI counting)
• Load-bearing user journey
• No existing scenario covers it (numbered 001/002/003)
• Concrete regression risk tied to a prior incident

Out of scope

The other 5 items in the task-page acceptance suite (mode coverage, per-language routing, context loading, citation discipline, broken-YAML isolation) are better as unit/integration tests on the specific components — they don't clear the four-criteria gate for an E2E scenario.

Test plan

• `make precommit` clean (131 rules, 28 mechanical YAMLs, no drift)
• Each scenario audited by `dark-factory:scenario-auditor` (twice)
• All audit-flagged issues resolved (placeholders → concrete commands, multi-mode → split, internal-string greps → contract-level assertions, ambiguous counter → PATH-shim with positive control)
• Bot review
• Walk scenario 001 manually to validate the dispatcher fail-fast (after merge)

[ben-s-pull-request-reviewer]

Reviewed by ben-s-pull-request-reviewer[bot] — no concerns flagged.

[ben-s-pull-request-reviewer]

{
  "verdict": "approve",
  "summary": "PR #37 has already been merged into master (commit 3b6adf2). The diff against master is empty — no reviewable changes remain. Manual review of the three new scenario files (001-toolchain-preflight.md, 002-clean-pr-zero-findings.md, 003-scaling-funnel-100-files.md) found two minor observations: Scenario 003 lacks a `trap` to guarantee `$WORK` cleanup on early-exit failure paths (the Cleanup step won't run if the scenario errors out midway), and all three scenarios remain `status: draft` with no documented promotion path to `status: accepted` if that's a project convention requirement.",
  "comments": [
    {
      "file": "scenarios/003-scaling-funnel-100-files.md",
      "line": 52,
      "severity": "minor",
      "message": "Should Fix (Important): The Cleanup step (line 69) runs only on success. If the scenario exits early due to an error or unmet assertion, `rm -rf \"$WORK\"` never executes and the mktemp directory leaks. Adding `trap 'rm -rf \"$WORK\"' EXIT` near the top of the Action block would guarantee cleanup regardless of exit path."
    },
    {
      "file": "scenarios/001-toolchain-preflight.md",
      "line": 13,
      "severity": "nit",
      "message": "Nice to Have (Optional): The `command -v ast-grep` check (line 17-18) validates PATH resolution but doesn't confirm the binary is actually executable. Consider adding `[ -x \"$(command -v ast-grep)\" ]` to verify not just visibility but executability in environments where PATH contains non-executable scripts with the same name."
    }
  ],
  "concerns_addressed": [
    "performance: Scenario 003 shim wrapper is reliable for session-scoped Claude CLI invocations — bypass is a known limitation documented in the scenario",
    "correctness: Scenario 001 dispatcher PATH preflight is validated — the expected `ast-grep/sg not in PATH` error message is specified and verifiable",
    "correctness: Scenario 002 report format uses literal `None.` strings — fragile by design (explicit trade-off documented), verified by specific awk extraction commands",
    "security: Scenario 003 cleanup runs on success but not on early-exit failure — see finding above",
    "tests: All scenarios have status: draft — no promotion criteria documented; may be intentional for test-contract documentation"
  ]
}