Implement BM25, Boolean, phrase, prefix, fuzzy, autocomplete, and highlighting queries

[kylebernhardy]

Outcome

Implement the agreed English full-text query behavior using Tantivy query primitives.

Acceptance criteria

• Weighted BM25 uses documented defaults and per-field boosts.
• Term-any, term-all, phrase, bounded prefix/autocomplete, bounded fuzzy, and preview fuzzy-prefix are supported.
• Structured candidate filters use a score-neutral boosted TermSetQuery.
• Fuzzy exact preference uses an exact scoring branch plus a lower constant fuzzy branch without double-counting BM25.
• Positions and surface terms are configurable capabilities.
• Highlighting/snippets are configurable and disabled by default.
• Expansion, candidate, clause, deadline, and result-memory ceilings are enforced.
• Stable score ordering and pagination behavior are covered by fixtures.

Native storage scope — September 14, 2026

Storage is native Tantivy MmapDirectory for standalone and Harper use. The query, ranking and capability acceptance criteria above remain unchanged; there is no hosted storage implementation to qualify.

Current design: Native Tantivy storage and Harper derived indexes.

Comment generated by kAIle (GPT-6)