refactor: switch Mumble voice to pymumble transport

asyncio's SSL memory-BIO transport silently drops voice packets even
though text works fine. pymumble uses blocking ssl.SSLSocket.send()
which reliably delivers voice data.

- Rewrite MumbleBot to use pymumble for connection, SSL, ping, and
  voice encoding/sending
- Bridge pymumble thread callbacks to asyncio via
  run_coroutine_threadsafe for text dispatch
- Voice via sound_output.add_sound(pcm) -- pymumble handles Opus
  encoding, packetization, and timing
- Remove custom protobuf codec, voice varint, and opus ctypes wrapper
- Add container patches for pymumble ssl.wrap_socket (Python 3.13) and
  opuslib find_library (musl/Alpine)

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

plugins/music.py

@@ -252,6 +252,17 @@ async def cmd_np(bot, message):
     )
 
 
+@command("testtone", help="Music: !testtone -- debug sine wave")
+async def cmd_testtone(bot, message):
+    """Send a 3-second test tone for voice debugging."""
+    if not _is_mumble(bot):
+        await bot.reply(message, "Mumble-only feature")
+        return
+    await bot.reply(message, "Sending 440Hz test tone (3s)...")
+    await bot.test_tone(3.0)
+    await bot.reply(message, "Test tone complete")
+
+
 @command("volume", help="Music: !volume [0-100]")
 async def cmd_volume(bot, message):
     """Get or set playback volume.