fix(roda-mcp-server): rank search_datasets by relevance, drop stop words

[ariharrisonlab]

Fixes #4462

Summary

search_datasets matched query terms by substring (only data/dataset/datasets were filtered) and then ordered results by provider diversity with no relevance score. Broad natural-language queries therefore matched hundreds of datasets, and the most relevant one could be pushed out of the top results by less-relevant datasets from other providers — so a more specific query returned worse results than a single keyword.

Changes

• Expand stop-word removal so multi-word queries narrow the result set instead of matching nearly everything via substring.
• Add a field-weighted relevance score (exact tag > name > description, plus a coverage bonus for matching more of the distinct query terms) and sort matches by score before the provider-diversification pass, so diversification applies within relevance order.
• Keep the permissive substring match gate (no recall regression) and the JSON response shape unchanged; the internal score is stripped from output.
• Add tests for relevance ordering, stop-word handling, and score omission.

User experience

Against the live registry index (1,177 datasets):

query	before (matches / best-match rank)	after (matches / best-match rank)
ADMET	1 / #1	1 / #1
cheminformatics	1 / #1	1 / #1
drug discovery molecules with ADMET predictions	746 / not in top 10	281 / #1

Before, adding descriptive words to a query buried the best match; after, a more specific query returns more relevant, better-ranked results (for that example the top 3 are the three actual drug-discovery datasets in the registry). Single-keyword and filtered-query behavior is unchanged.

Checklist

• I have reviewed the contributing guidelines
• I have performed a self-review of this change
• Changes have been tested (pytest tests/ — 93 passed; ruff check + ruff format clean)
• Changes are documented (code comments + tests)

Is this a breaking change? (N)

Acknowledgment

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of the project license.