Match Devanagari names in the member directory search

[burakeyler]

What changed

Fixes #5. The member directory search now normalizes both the query and the member fields to Unicode NFC before matching, so a Nepali (Devanagari) name matches whether it was stored or typed in precomposed or decomposed form (e.g. ऩ U+0929 vs न + nukta U+0928 U+093C). Previously the plain toLowerCase().includes treated those as different strings.

The filter moved out of members/page.tsx into src/server/members/search.ts (memberMatchesQuery) so it can be unit tested without a database; the page behaviour is otherwise unchanged (same fields, trimmed, case-insensitive, empty query matches all).

Contract impact

• No changes to packages/shared
• packages/shared changed — needs a reviewer from the frontend and the backend

Checks

• bun run lint — ran Biome 2.5.13 on the changed files; clean
• bun run typecheck — not run locally (see notes)
• bun run test — new tests/unit/member-search.test.ts passes (6 tests) with Vitest 5
• bun run build and/or bun run build:web

Notes for reviewers

• I don't have bun/Postgres set up on this Windows machine, so I ran the new unit test with a DB-less Vitest config and did not run typecheck, the integration suite, or the build. CI should cover those; happy to fix anything it flags.
• Tests cover a Devanagari display-name match, decomposed query vs precomposed name and the reverse, a non-match, Latin case-insensitivity/trimming, and a null headline.

🤖 Generated with Claude Code