Handle invalid UTF-8 in whisper transcription responses

[74th]

Summary

• decode whisper.cpp and whisper-server JSON responses with UTF-8 replacement so malformed byte sequences do not abort transcription loading
• emit warnings when replacement characters are introduced while parsing transcript JSON
• add unit tests covering malformed UTF-8 handling for both whisper.cpp and whisper-server paths

Testing

• python -m unittest discover -s tests -p 'test_whisper_server.py'
• make lint