Add HTTP service, MCP adapter, systemd autostart; fix bugs and docs

- chatterbox_cli_v4.py: cooperative stop/interrupt via threading.Event;
  fix force_split_sentence (word boundary instead of mid-word cut);
  fix synthesize_streaming normalization order (split before preprocess)
- tts_service.py: FastAPI service with job queue, model cache, worker thread;
  LAN-accessible on 0.0.0.0:9999; audio_device default None (auto)
- mcp_adapter.py: MCP adapter (stdio + streamable-http) wrapping REST API;
  update docstring and default TTS_URL to port 9999
- requirements.txt: add fastapi, uvicorn, httpx, mcp
- README.md, BEDIENUNGSANLEITUNG.md: document service, MCP, AI integrations
  (Claude, Ollama, Open WebUI, llama.cpp, Home Assistant), systemd autostart
- CLAUDE.md: reflect current architecture (service + adapter now implemented)

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

chatterbox_cli_v4.py

@@ -463,7 +463,9 @@ def force_split_sentence(text: str, max_len: int) -> List[str]:
     while len(remaining) > max_len:
         split_pos = remaining.rfind(" ", 0, max_len + 1)
         if split_pos <= 0:
-            split_pos = max_len
+            # Kein Leerzeichen vor max_len — vorwärts zum nächsten Wortende suchen
+            next_space = remaining.find(" ", max_len)
+            split_pos = next_space if next_space != -1 else len(remaining)
         parts.append(remaining[:split_pos].strip())
         remaining = remaining[split_pos:].strip()
 
@@ -830,15 +832,9 @@ def synthesize_streaming(
     if voice_path and not Path(voice_path).exists():
         raise FileNotFoundError(f"Voice-Referenz nicht gefunden: {voice_path}")
 
-    text = preprocess_tts_text(
-        text=text,
-        lang=lang,
-        spell_uppercase_acronyms=spell_uppercase_acronyms,
-        acronym_mode=acronym_mode,
-        normalize_time_values=normalize_time_values,
-        normalize_year_values=normalize_year_values,
-        normalize_units_values=normalize_units_values,
-    )
+    # Erst bereinigen und splitten, dann pro Chunk normalisieren —
+    # sonst erzeugen Akronym-Punkte ("ARD" → "Ah Er De.") falsche Satzgrenzen.
+    text = clean_raw_text(text)
 
     model, model_kind, sr = load_model(lang, device)
 
@@ -849,9 +845,20 @@ def synthesize_streaming(
         )
 
     if conversation_mode:
-        text_chunks = split_for_conversation(text, first_chunk_len=first_chunk_len, max_len=max_len)
+        raw_chunks = split_for_conversation(text, first_chunk_len=first_chunk_len, max_len=max_len)
     else:
-        text_chunks = split_long_text(text, max_len=max_len)
+        raw_chunks = split_long_text(text, max_len=max_len)
+
+    preprocess_kw = dict(
+        lang=lang,
+        spell_uppercase_acronyms=spell_uppercase_acronyms,
+        acronym_mode=acronym_mode,
+        normalize_time_values=normalize_time_values,
+        normalize_year_values=normalize_year_values,
+        normalize_units_values=normalize_units_values,
+    )
+    text_chunks = [preprocess_tts_text(c, **preprocess_kw) for c in raw_chunks]
+    text_chunks = [c for c in text_chunks if c.strip()]
 
     if not text_chunks:
         raise ValueError("Kein verwertbarer Text nach dem Einlesen gefunden.")