feat(mcp): add reusable evidence boards

[tulayha]

Summary

• return an annotated evidence board in the ordinary completed search_moments and query_video job
• board the ranked result set without the five-item standalone frame/clip limit; preserve next_start_rank when the page budget is exceeded
• keep keyframes and keyframes_and_clips as optional same-job drill-down artifacts
• keep create_evidence_board only for custom selections and continuation pages
• use the same board contract over local stdio and Streamable HTTP MCP, including inline images, resource links, and tile-to-evidence mapping
• simplify the public ingestion/evidence skills and align the README and architecture documentation with the actual default flow

Why

The intended agent path is one retrieval submission and one polled job that already contains a visual overview. Requiring a second board job—or forcing the agent through repeated batches of individual frames—defeated that purpose.

Agent flow

1. Submit search_moments or query_video without command.evidence_delivery.
2. Poll that job with get_job.
3. Inspect the returned board and its stable evidence IDs.
4. Materialize only selected IDs as standalone keyframes or clips when needed.

Validation

• 135 focused application, evidence, MCP, API, and projection tests passed with 7 subtests
• real official stdio MCP client: omitted evidence options, one search job succeeded with one inline JPEG board, one ResourceLink, six ranked tiles, and zero standalone items
• both repo and installed global skills pass quick_validate.py and are byte-identical
• Ruff and git diff --check passed

No migration, dependency, frontend, actor behavior, or captioning change is included.

BEGIN_COMMIT_OVERRIDE
feat(mcp): return visual evidence boards with search results
END_COMMIT_OVERRIDE