[P2] Build Code2TestBench harness + validate metrics

[pro-utkarshM]

Problem

Proposal §7 promises Code2TestBench and headline metrics (70% acceptance, 80% first-run pass). No benchmark harness exists; metrics are unvalidated.

Tasks

• Harness: run generation against sample repos (flask, requests) with their real tests hidden.
• Measure acceptance rate, first-run pass rate, diagnostic accuracy.
• Record results in docs; reconcile with proposal targets.

Acceptance

Reproducible benchmark command + a results table committed.