Support for Batched inference and language detection from multiple segments in faster-whisper

[Jiltseb]

This PR adds below capabilities for the faster-whisper project:

1. Better language prediction using multiple audio segments.
2. Batched inference for faster transcription.
3. Streaming (segment-level) or non-streaming options for Batched inference.

This will be used as the faster-whisper package for Aana SDK that supports Batched Inference.

[evanderiel]

I see you've removed the suppress_numerals functionality entirely. Is there a reason for that?

[Jiltseb]

This function was added as a means to suppress numeric symbols and currency symbols. But this will be used only if we are using a separate ASR model for alignment purposes and that model can not align numerals. Actually, I can avoid it for whisper transcription. Thanks for pointing this out.

I see you've removed the suppress_numerals functionality entirely. Is there a reason for that?