Add structured commissioner scores

[KyleHerndon]

Summary

• Add a structured scores map to commissioner episode scores while preserving the legacy scalar score field.
• Propagate structured score maps through the round-start adapter into local commissioner models.
• Aggregate named score fields into ranking.result_metadata["scores"] at round completion so app_backend can publish multiple leaderboard score columns.

Validation

• uv run python -m compileall commissioners tests — passed.
• uv run pytest tests/test_commissioner_strategies.py -q — 71 passed.
• git diff --check — passed.

Lint

• No lint command is configured in pyproject.toml; the dev dependency group contains pytest/pytest-asyncio/httpx only.