Speech to text improvements

[andremion]

Goal

Refactor speech-to-text implementation to improve state management, API design, and code organization, and fix a bug when the recording is manually stopped, and the previously recognized text is partially lost.

Implementation

• Refactored SpeechToTextButtonState to use Compose state properties (isListening, rmsdB) and moved onTextRecognized callback to state creation level
• Simplified SpeechRecognizerHelper interface by removing multiple callbacks and using properties instead of methods
• Removed unnecessary ComponentActivity dependency and DisposableEffect from SpeechToTextButton
• Extracted helper functions and improved state reset logic

Out of scope

• Updated sample app credentials

Testing

• Start the recognizer, speak, see the recognized text, and stop manually.
• Repeat the steps above, but let the recognizer stop automatically after 3 seconds of silence.

Checklist

• Issue linked (if any)
• Tests/docs updated
• I have signed the Stream CLA (required for external contributors)

[github-actions]

SDK Size Comparison 📏

SDK	Before	After	Difference	Status
stream-chat-android-ai-compose	0.01 MB	0.01 MB	0.00 MB	🟢

[andremion]

redundant setup