[codex] v3.29 OpenAI voice command input

[Anthonyiswhy]

Summary

Adds the v3.29 push-to-talk voice command layer without changing the safety alert TTS output path.

• Adds optional BLINDNAV_VOICE_INPUT=1 command capture with arecord
• Transcribes command WAVs through OpenAI STT using gpt-4o-mini-transcribe by default
• Routes deterministic commands: describe, nearest, people, status, repeat, cancel
• Keeps urgent/warning obstacle alerts on the existing local VoiceAssistant/Piper path
• Adds tools/test_transcribe.py for command-line STT smoke testing

Validation

• python -m py_compile raspberry_pi\yolo_realsense_navigation.py tests\test_blindnav_v326.py tools\test_transcribe.py
• python -m pytest tests/test_blindnav.py tests/test_blindnav_v326.py -q -> 192 passed in 10.17s

Notes

This PR is the voice-input-only branch. It does not include the v3.30 OpenAI alert TTS field-test mode.