fix(tokenizer): block kanji reading collisions and name/generic length t

- Add allowReadingOnlyMatch:false to kanji token known-word lookups so 渓谷/けいこく no longer matches a mined 警告/けいこく card via reading-only index
- Greedy name pre-pass yields when a strictly longer generic word starts at the same position (空 no longer splits 空気; ties still go to the name)

src/anki-integration/known-word-cache.ts

@@ -142,7 +142,11 @@ export class KnownWordCacheManager {
     );
   }
 
-  isKnownWord(text: string, reading?: string): boolean {
+  isKnownWord(
+    text: string,
+    reading?: string,
+    options?: { allowReadingOnlyMatch?: boolean },
+  ): boolean {
     if (!this.isKnownWordCacheEnabled()) {
       return false;
     }
@@ -163,6 +167,14 @@ export class KnownWordCacheManager {
       );
     }
 
+    // Callers that look up a kanji token's reading (not subtitle text) must
+    // opt out of the reading-only fallback: readingCounts holds readings of
+    // every note including kanji words, so 渓谷's けいこく would match a
+    // mined 警告/けいこく.
+    if (options?.allowReadingOnlyMatch === false) {
+      return false;
+    }
+
     // Reading-only fallback, except for single-kana text: particles and
     // interjections (よ, ね, え…) would otherwise borrow the reading of an
     // unrelated note (夜「よ」, 絵「え」) and count as known.