feat: wire BM25F cross-field aggregation with configurable per-field weights

[nnunley]

Summary

• Introduce QueryNode::TermExpansion to preserve the full searched default field set separately from resolved term children, enabling aggregate scoring
• Wire BM25F cross-field aggregation: when execution sees a TermExpansion under a BM25F scorer, it collects per-document field stats (including zero-frequency entries for absent fields) into a single Bm25FScorer::score call
• Add configurable per-field weights via PlanningContext::with_field_weights; weights flow through TermExpansion into FieldStats.weight. Fields not explicitly weighted default to 1.0
• Explicit boolean OR remains a sum-of-child-scores operator; only planner-generated default-field expansions use aggregate scoring

Test plan

• All existing tests pass (cargo test --workspace)
• New tests in search_behavior.rs verify:
  • BM25F aggregate score matches hand-computed Bm25FScorer::score output
  • Unique document frequency across fields produces valid scores
  • Default boost multiplies final aggregate score exactly once
  • Duplicate same-field OR falls back to generic OR summing
  • Explicit cross-field OR uses generic OR (not BM25F aggregation)
  • Non-hit field lengths affect aggregate scores
  • Body lengths affect BM25F even when term resolves only in title
  • Field weights affect ranking (title-heavy weights rank title matches first)
  • Exact-score verification with weighted Bm25FScorer::score output
• Planner tests verify TermExpansion carries field weights
• Pre-push hooks pass (clippy, fmt, doc, no_std, wasm32)

🤖 Generated with Claude Code

[waywardmonkeys]

LGTM!