fix(worker): filter the transliterated 'fuer' stopword, not the untransliterated 'fur'

NormalizeTitleWords transliterates umlauts before tokenizing ('für' ->
'fuer'), but TitleStopWords listed 'fur' -- a token the pipeline can
never produce -- so 'für' was never filtered as filler and could push
unrelated titles past the near-duplicate threshold.

src/ClaudeDo.Worker/External/ExternalMcpService.cs

@@ -329,7 +329,7 @@ public sealed class ExternalMcpService
     // plus the recurring "mcp"/"task(s)"/"cleanup" nouns called out in the task write-up).
     private static readonly HashSet<string> TitleStopWords = new(StringComparer.Ordinal)
     {
-        "der", "die", "das", "und", "von", "auf", "mit", "fur", "ein", "eine",
+        "der", "die", "das", "und", "von", "auf", "mit", "fuer", "ein", "eine",
         "ist", "sind", "oder", "nicht", "mcp", "task", "tasks", "cleanup",
     };