feat: add Hebrew proximity wildcard search

[david-hoze]

Summary

• Adds a new @N(pattern1 pattern2 ...) search syntax that finds Hebrew words matching wildcard patterns within N words of each other (N is optional, defaults to 10)
• Supports * (any Hebrew letter), + (prefix/suffix letter set), final-form interchangeability (e.g. מ/ם, נ/ן), and automatic nikud stripping
• Scans all indexed documents directly for proximity queries, bypassing MiniSearch pre-filtering (Hebrew prefix letters make token-based pre-filtering unreliable)
• Handles both LTR (@10(תקון)) and RTL ()10(תקון@) input forms

Files changed

File	Action
src/search/proximity-search.ts	Created — core proximity search logic (parsing, regex compilation, tokenizer, proximity window finder, match conversion)
src/search/query.ts	Modified — detect @N(...) syntax early in the Query constructor, store parsed ProximityQuery, populate root terms for downstream use
src/search/search-engine.ts	Modified — added getProximitySuggestions() that scans all indexed documents with proximity matching, called from getSuggestions() when a proximity query is detected
src/__tests__/proximity-search-tests.ts	Created — unit tests for parsing, regex compilation, tokenization, proximity matching, and root term extraction

Test plan

• Open Omnisearch (Vault search) and enter @10(תקון) — should return notes containing the Hebrew word תקון (and variants like תִּקּוּן with nikud)
• Verify highlighting works on matched words in the result excerpts
• Test multi-pattern query @10(+ת*ק*ן+ +ש+למ+ מאד) — should find documents where all three patterns appear within 10 Hebrew words of each other
• Test single-pattern query @(+תקן+) (no distance number) — should default to distance 10
• Test RTL input form )10(תקון@ — should be recognized and work identically
• Verify that normal (non-proximity) searches are completely unaffected
• Run existing test suite to confirm no regressions