fix(anki): regenerate sentence furigana when mined context changes

- Regenerate SentenceFurigana from the final sentence after timing-review expansion or stats-dashboard word mining
- Add a Yomitan parseText-based generator with a 10s timeout and escaped, highlighted output
- Clear stale furigana on failure so templates fall back to Sentence
- Keep existing furigana formatting when the sentence is unchanged
- Document the behavior in the anki-integration and immersion-tracking docs

src/main/runtime/stats-server-runtime.ts

@@ -1,3 +1,4 @@
+import { generateSentenceFurigana } from '../../core/services/tokenizer/sentence-furigana';
 import path from 'node:path';
 import type { BrowserWindow } from 'electron';
 import {
@@ -169,6 +170,8 @@ export function createStatsServerRuntime(deps: StatsServerRuntimeDeps): {
       }),
       resolveAnkiNoteId: (noteId: number) => deps.resolveAnkiNoteId(noteId),
       resolveSentenceSearchHeadwords: (term: string) => deps.resolveSentenceSearchHeadwords(term),
+      generateSentenceFurigana: (text, highlightedText) =>
+        generateSentenceFurigana(text, highlightedText, yomitanDeps, yomitanLogger),
       addYomitanNote: async (word: string) => {
         const ankiConnectConfig = deps.getResolvedConfig().ankiConnect;
         const ankiUrl = getPreferredYomitanAnkiServerUrl(ankiConnectConfig);