chore: evaluate Fuse.js to replace custom fuzzy search implementation

Summary

src/utils/searchUtils.ts contains a hand-rolled Levenshtein distance algorithm with a weighted scoring system (exact: 100, substring: 60, fuzzy: 40). This works for current dataset sizes but carries maintenance burden and edge-case risk.

Candidate Library

Fuse.js — https://www.fusejs.io/

• ~4KB gzipped
• Apache-2.0 license
• Battle-tested fuzzy search with ranking
• Configurable thresholds and key weights

Action

Benchmark Fuse.js against the current implementation on the actual patient/doctor datasets. If relevance quality is equal or better, replace searchUtils.ts (~47 lines) with a thin wrapper around Fuse.js.

If datasets remain small (<500 items) and current results are acceptable, close this issue and document the tradeoff in a code comment.

Acceptance Criteria

• Benchmark run on real data
• Decision documented (replace or keep with comment)

Labels: chore, dependencies, performance