perf(vector/index): eliminate `field_index: HashMap<String, ...>` lookup hash on every NRT query

[mosuka]

Round-3 perf push sub-issue (tracked under umbrella #535).

[S] Eliminate field_index: HashMap<String, ...> lookup hash on every NRT query

• Where: laurus/src/vector/index/quantized_storage.rs:127-129
  (field_position_index clones an Arc but the first lookup still
  hashes the String); same in pq_storage.rs:113-115.
• Current behaviour: NRT path hashes the field name string per
  query; with sub-microsecond distance kernels, this is now a measurable
  fraction.
• Suggested direction: cache the (field_name, field_id) mapping
  in the searcher itself (one lookup at HnswSearcher / IvfSearcher
  construction). Pair with Fix fuzzy query example #8 + Fix wildcard query #14 to make the field_id the only
  identity carried inside the pool.
• Risk / scope: trivial.

---

---

ID: VI-24 — see ~/.claude/tasks/laurus/20260523_perf_round3_audit/task_list.md for the full Round-3 issue list.

[mosuka]

Batch re-triage of the Round-3 audit backlog (2026-08-02) — premise partially overtaken; remaining scope: the "per query" framing is stale — field_position_index is resolved once per search per segment pool, not per node, leaving O(segments) String hashes per query. Remaining: key the pools by the #633 u16 field id so no String identity is carried at all; bench-gated (the residual cost is small).