The scorecard splits each crew member's figures by the model and prompt that ran #3169

Part of #2677: 'start tracking which crew members are better'. Runs record model and prompt (migration 20261019130000-run-model-prompt).

Done:

• [ ] The scorecard can group by (member, model, prompt variant); the crew page shows one sub-row per pairing a member used in the window, with the same figures.
• [ ] A test seeds two models for one member and gets two rows with their own landed counts.

Files: src/core/scorecard.ts, src/web/pages/crew.ts, a test. Risk: pulls opened before the migration have no model; show them as 'unrecorded', never drop them.