feat: add voice input mode via /speak

[ercbot]

Resolves #418

Adds a /speak command in the CLUI which triggers a transcription based input to the text box

Uses the OpenAI Realtime API for transcription of voice input

When /speak is invoked:

• A red ● recording indicator appears at the front of the input box
• Audio is streamed to the OpenAI api and a VAD detects when you start speaking, once it detects you have finished speaking the transcription is streamed from the transcription model to the input box.
• Pressing enter sends the recorded text, and pauses any transcription until the input box opens again
• Pressing any other key automatically stops the recording so you can make edits (useful if you're being more precise and editing your responses or in a noisy environment for instant)

The default language is 'en' and the default model is 'gpt-4o-mini-transcribe'

I was originally going to add language config to the command itself but it might be better suited for the ~/.codex/ config file so I am looking for feedback on the best place to add config for model + language.

Tested and working Macbook Air M4 with built in microphone

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[ercbot]

I have read the CLA Document and I hereby sign the CLA

[ercbot]

I addressed all feedback from the previous review and added tests for both the transcriber.ts module and the terminal-chat-input.tsx component to cover the new transcription functionality.

I also added config settings to customize the transcription model and language through the existing config system. Details for these options are available in the OpenAI docs.

Functionally, everything is working as intended. Let me know if there's anything else needed to complete this PR.

[ercbot]

Example of the .codex/config.yaml file:

# existing config options
model: 4o-mini
approvalMode: suggest 
# ...
# new transcription options (these are the defaults found in src/utils/transcriber.ts)
transcription:
  input_audio_transcription:
    model: gpt-4o-transcribe
    prompt: ""
    language: "en"
  turn_detection:
    type: server_vad
    threshold: 0.6
    prefix_padding_ms: 400
    silence_duration_ms: 500
  input_audio_noise_reduction:
    type: near_field