feat: replace wanakana with english-to-kana 49K dictionary

- Vendor english-kana-matcher.js (english-to-kana, MIT, 49216 words)
- Rewrite bridge: dictionary lookup + letter-by-letter fallback for OOV words
- Convert English punctuation (,.!?) to Japanese pauses (、。)
- Fix Chinese fullwidth comma being stripped by filter_kana
- Preserve decimal points in numbers (3.5 -> スリーファイブ)
- main.py now applies filter_kana (matches web UI behavior)
- Remove wanakana dependency; rebuild portable zip (82.5 MB)

main.py

@@ -12,7 +12,7 @@ import signal
 import sys
 
 from danmaku_handler import DanmakuClient
-from chinese2kana import chinese_to_kana
+from chinese2kana import chinese_to_kana, filter_kana
 from tts import TTSBridge
 from audio_player import play_wav_async, get_output_devices
 
@@ -47,6 +47,7 @@ async def tts_worker(queue: asyncio.Queue, bridge: TTSBridge,
             continue
         try:
             kana = chinese_to_kana(text, convert_numbers=convert_numbers)
+            kana = filter_kana(kana)
             logger.info("Speaking: %s -> %s", text, kana)
             wav_data = await bridge.synthesize(kana)
             await play_wav_async(wav_data)