Voice input

• Set `openAiApiKey` in app settings before voice input is enabled.
• Microphone permission must be granted by your OS/browser runtime.
• Transcription runs through OpenAI transcription model `gpt-4o-transcribe` in current implementation.

• Start recording from the mic button in the input action row.
• When stopping, the flow captures trailing audio, then submits the blob for transcription.
• On success, transcript is either appended to input or auto-sent, depending on settings.
• On failure, a transcription error notification is shown instead of silently dropping input.

If `autoSendVoiceMessages` is enabled and your text input is empty, the transcript is sent immediately. Otherwise it is appended to current draft text.

• Use auto-send for short command-style prompts.
• Keep auto-send off when you prefer to edit transcriptions before execution.
• Use a headset in noisy environments to reduce transcript cleanup.