Investigate on a better gpu usage

[chainyo]

This PR proposes an improvement to transcription by implementing a batch request process.

• Add README instructions for profiling the container
• Fix Exception/Error returns through the API -> raised errors should be more transparent for the user
• VAD use now onnx and faster-whisper implementation
• Transcription is now batched. Current batch size of 32 (less than 13GB of VRAM) -> TODO: make it configurable based on the GPU space.
• word_timestamps is disabled for now until it's fully implemented with the batched transcription process