Expand weak cross-modal benchmark categories

[brianmeyer]

Summary

• Expand the REC-168 weak media-query categories to 20 queries each: image_to_text, image_to_document, video_to_text, video_to_image, and video_to_document.
• Add grounded media-query variants with explicit source-path provenance and graded relevance where applicable.
• Add regression coverage for the weak-category query floor and media source path preservation.
• Update the UAT corpus benchmark distribution docs to 231 total queries and document parent-memory vs asset-level metrics.

Research grounding

• BEIR-style qrels/results metrics: https://github.com/beir-cellar/beir
• MTEB retrieval evaluation framework: https://github.com/embeddings-benchmark/mteb
• ViDoRe pipeline evaluation patterns: https://github.com/illuin-tech/vidore-benchmark

Tests

• python3 benchmarks/cross_modal_ablation.py --dry-run
• python3 -m pytest -q tests/test_cross_modal_benchmark_defs.py tests/test_cross_modal_diagnostics.py
• python3 -m pytest -q