Refactor speaker recognition client to use in-memory audio data

- Updated methods to accept audio data as bytes instead of file paths, enhancing performance by eliminating disk I/O.
- Improved logging to reflect in-memory audio processing, providing better insights during speaker identification and diarization.
- Streamlined audio data handling in the `diarize_identify_match` and `diarize_and_identify` methods, ensuring consistency across the client.
- Removed temporary file handling, simplifying the audio processing workflow and reducing potential file system errors.

Author: AnkushMalaker