feat(voice): transcribe Telegram voice notes with SpeechRecognition

Add a voice message handler that downloads OGG audio, converts it to
WAV via pydub/ffmpeg, and transcribes it with Google Speech Recognition
(pt-BR) before feeding the text into the normal message pipeline.

Also extract shared text processing into process_text_message, guard
against updates without a text attribute, switch the default AI provider
to openrouter with deepseek-v4-flash, and document the ffmpeg system
requirement in DEPLOY.md.

venv/lib/python3.11/site-packages/pydub/logging_utils.py

@@ -0,0 +1,14 @@
+"""
+
+"""
+import logging
+
+converter_logger = logging.getLogger("pydub.converter")
+
+def log_conversion(conversion_command):
+    converter_logger.debug("subprocess.call(%s)", repr(conversion_command))
+
+def log_subprocess_output(output):
+    if output:
+        for line in output.rstrip().splitlines():
+            converter_logger.debug('subprocess output: %s', line.rstrip())