Implement `only_transcription` and `only_diarization`

[chainyo]

This PR:

• Simplify the endpoints added to the asr_type
• Added the transcription_endpoint and diarization_endpoint
• Fix the dependencies of the project by moving inference to core dependencies.
• Define the ASRTranscriptionOnly for the only_transcription asr type.
• Define the ASRDiarizationOnly for the only_diarization asr type.
• Removed the clueless Word model to use the one from faster-whisper.
• Updated how endpoints are added.
• Updated the TranscribeService to handle audio formatted as TensorShare object for remote execution case.
• Corrected the read_audio method to raise a ValueError if the input type is not matched.

Fixes #259 and #260