feat(voice): Pas 2 — install voice deps, vendor discord-ext-voice-recv, setup assets

Foundation pentru Discord voice-to-voice pipeline.

- requirements.txt: faster-whisper, silero-vad, num2words, numpy, PyNaCl
- vendor/discord-ext-voice-recv/: vendored la commit ac04ea7b09 (bump version
  0.5.3a) — Discord voice protocol fragil, upstream hobby fork. Adapter layer
  in src/voice/_discord_voice_adapter.py izolează churn (swap la py-cord =
  doar acel fișier rescris). VENDOR_INFO.md documentează update procedure.
- tools/voice_setup.py: idempotent setup script — libopus check, ffmpeg
  check, Supertonic reachable, faster-whisper/silero-vad warm, assets
  generation. Exit 0 = green, 1 = needs human (currently libopus missing
  needs `sudo apt install -y libopus0`).
- assets/voice/: thinking.wav (filler "Stai puțin să-mi adun gândurile",
  ~2.8s), mhm.wav (listener noise), beep_200ms.wav (wake-up tone 880Hz).
- src/voice/__init__.py: package stub.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

vendor/discord-ext-voice-recv/update_notes.md

@@ -0,0 +1,21 @@
+# Update notes
+Notably, not a changelog, just notes.
+
+## 0.5.2
+- Adds `extras.localplayback` module
+- Adds info about the extras modules to the readme
+- Adds `WavSink` as an alias to `WaveSink`
+- Fixed a member cleanup error in SpeechRecognitionSink
+- Changes the optional dependency format
+  - Previously it was a single optional dep, `extras`.  Now there is a dependency per module, with `extras` installing all of them.  See the readme for details.
+
+## 0.5.1
+- Fixes a build process related error
+- Changes `voice_recv.extras` import semantics
+  - The `__all__` contents of the extras modules are no longer `*` imported into `voice_recv.extras` (this was only `extras.SpeechRecognitionSink`).  You will have to access them directly, or import that specific extra module.  Example:
+    ```py
+    from discord.ext.voice_recv.extras.speechrecognition import SpeechRecognitionSink
+    # or
+    from discord.ext.voice_recv.extras import speechrecognition
+    sink = speechrecognition.SpeechRecognitionSink(...)
+    ```