Expand Memory.swift evals with metadata-aware smoke coverage and grouped gates

[zac]

Summary

• Added shared eval row metadata and stricter dataset validation across storage, recall, query expansion, and agent-memory suites.
• Introduced a small committed synthetic smoke dataset plus a focused baseline for category-aware gating.
• Updated reports, reduced metrics, and Markdown output to surface grouped metrics by category, source family, and difficulty.
• Refined memory extraction, debug, storage migration, and CoreML default configuration behavior to support the new eval coverage.
• Added/updated reusable skills for designing memory evals and synthetic datasets.

Testing

• Added and updated unit coverage for the new metadata, migration, and eval-reporting paths.
• Validated the updated eval dataset structure and baseline shape.
• Not run (not requested)