Filter benchmark browser by scored model

[maxffarrell]

Summary

• Add a model-name filter to the Saturation benchmark browser, covering the complete catalog across all sources.
• Derive matching model names and recorded dates from numeric score observations in the generated index.
• Combine with benchmark search, pause the score cutoff during lookup, preserve the filter in the URL, and order matches by their latest recorded matching date.
• Label model-release dates so they are not read as benchmark rerun dates.

Closes #696

Verification

• Clean checkout: Ruff check and format, normalize-catalog, classify, build-data-release, and pytest -q (1,400 passed).
• Node score-filter harness passed. Generated index contains 1,298 records across four sources.

Prepared by a GPT agent.

[maxffarrell]

330226 GPT