Add BC PR-Review engine arm with BCQuality ref override

[gggdttt]

What

Lands the BC PR-Review engine arm (faithful convergence arm) in BC-Bench and wires it into the CLI.

The engine (microsoft/BC-ALReviewAgent) now owns orchestration + the default BCQuality policy. This PR makes BC-Bench a first-class consumer of that engine, running its production Invoke-CopilotPRReview.ps1 in local mode as the review agent and scoring the result with BC-Bench's own CodeReviewPipeline.

Changes

• src/bcbench/agent/engine/ (new): the engine adapter
  • agent.py — run_engine_review(...): commits the patched worktree, fetches+filters BCQuality, invokes the engine, and normalizes al-code-review-findings.json into BC-Bench's flat review.json.
  • fetch_bcquality.ps1 — clones + layer-filters BCQuality at the configured ref.
• src/bcbench/agent/__init__.py — export run_engine_review.
• src/bcbench/commands/evaluate.py — new bcbench evaluate engine command.

BCQuality ref override (no engine edits)

run_engine_review accepts optional bcquality_repo / bcquality_ref, passed to the engine's fetch via the BCQUALITY_REPO / BCQUALITY_REF env vars (which the engine's Get-BCQualityConfig already honors on top of its baseline config). This lets a CI run review against a modified BCQuality branch/SHA without editing the engine.

Exposed on the CLI (all env-bound for Actions):

• --engine-scripts-dir (env ENGINE_SCRIPTS_DIR)
• --bcquality-repo (env BCQUALITY_REPO)
• --bcquality-ref (env BCQUALITY_REF)

Not in this PR (follow-ups)

• A GitHub Actions workflow entry to run the engine arm on CI.
• Recording the resolved BCQuality SHA + engine ref into ExperimentConfiguration for reproducibility.