Batch Processing not implemented for LlavaStreamGenerator

[rahulthakur319]

Currently, the LlavaStreamGenerator function at tinychat/stream_generators/llava_stream_gen.py processes inputs one at a time. To improve performance and throughput, we should implement batch processing capabilities. This will allow the function to handle multiple inputs simultaneously, potentially leading to significant speed improvements.

Quick action plan is to adjust the main generation loop to work with multiple sequences. But before starting wanted to confirm, Are there any model-specific considerations for batch processing with VILA models?