Add support for offline Silero VAD loading

[hungdoba]

Description

• Adds support for offline Silero VAD model loading by introducing silero_repo_or_dir, silero_source and silero_model parameters to AudioToTextRecorder. This allows users to specify a local directory or pre-loaded model, preventing torch.hub.load from attempting to download from GitHub in offline environments.
• The change also allows customization of the Silero VAD model name for future-proofing against new model variants.

Changes

• Added silero_repo_or_dir parameter to specify a local directory for Silero VAD.
• Added silero_source parameter to specify source of models, "local" or "github".
• Added silero_model parameter to accept a pre-loaded model.
• Updated torch.hub.load to use source="local" for local directories.
• Updated README.md with new parameter documentation.
• Added test script offline_vad.py to verify offline functionality.

Testing

• Tested with a local Silero VAD directory (./silero-vad) in an offline environment.
• Verified transcription works with silero_deactivity_detection=True.