Add backend tag boost for search without hard tag filtering

[bernoussama]

Problem

Agents should be able to provide tags as first-class search context, but simply concatenating tags into the search query can make keyword search worse. For example, appending typescript monorepo to a precise query like TS2307 pnpm may over-constrain websearch_to_tsquery and miss otherwise relevant results that match the error/package but not every tag.

Proposed approach

Keep query and tags separate in the API/search layer, then use tag overlap as a ranking boost instead of a required match.

Desired behavior:

• query remains the primary retrieval signal.
• tags are optional relevance hints.
• Results with overlapping tags rank higher.
• Results without tag overlap can still appear when they match the query well.
• No hard filtering unless explicitly added as a separate future option.

Implementation notes

• Extend search inputs to accept optional tags separately from query.
• Normalize tags consistently across stored solutions and input hints.
• For keyword search, add ranking boost for tag overlap without adding tag terms to the tsquery in a way that requires them.
• For hybrid/semantic search, preserve the query embedding behavior and apply tag overlap as a post-retrieval rerank signal.
• Update MCP search_solutions only after backend support exists, so agents can pass tags without degrading keyword recall.

Acceptance criteria

• Searching TS2307 pnpm with tags like typescript,monorepo does not lose strong TS2307 pnpm matches that lack one of those tags.
• Results with matching tags are ranked above similar-scoring results without matching tags.
• MCP/API docs describe tags as optional relevance hints, not strict filters.
• Tests cover keyword search with tags, tag-overlap reranking, and no-result regressions from over-constrained tag terms.