feat: romaji index for Japanese kanji song search

Backend:
- Add /api/romaji-index endpoint using kuromoji morphological analyzer
- Fetches all Navidrome songs, converts kanji → hiragana readings → romaji
- Only indexes songs with Japanese characters; caches for 20 min
- /api/romaji-index DELETE clears cache (force rebuild)

Frontend:
- fetchRomajiIndex() in api.js: lazy-fetches and caches the index
- doLocalSearch: when query is romaji (ASCII), queries romaji index in
  parallel with Navidrome search3 and merges+deduplicates results

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/api.js

@@ -39,6 +39,25 @@ export async function ndAuth() {
   return S._ndToken;
 }
 
+// ─── ROMAJI INDEX ───
+// Fetched once per session from the backend, cached in memory.
+let _romajiIndex = null;
+let _romajiIndexFetch = null; // in-flight promise dedup
+
+export async function fetchRomajiIndex() {
+  if (!S.cfg.backendUrl || !S.cfg.url || !S.cfg.user || !S.cfg.pass) return null;
+  if (_romajiIndex) return _romajiIndex;
+  if (_romajiIndexFetch) return _romajiIndexFetch;
+  _romajiIndexFetch = fetch(
+    `${S.cfg.backendUrl}/api/romaji-index?ndUrl=${encodeURIComponent(S.cfg.url)}&user=${encodeURIComponent(S.cfg.user)}&pass=${encodeURIComponent(S.cfg.pass)}`
+  ).then(r => r.ok ? r.json() : null)
+   .then(idx => { _romajiIndex = idx; _romajiIndexFetch = null; return idx; })
+   .catch(() => { _romajiIndexFetch = null; return null; });
+  return _romajiIndexFetch;
+}
+
+export function invalidateRomajiIndex() { _romajiIndex = null; }
+
 // ─── ITUNES ───
 // Route through backend proxy to avoid CORS; fall back to direct if no backend configured.
 function _itunesBase() {