added dockerfile and pip dependencies for OSD and LLM model compression workloads

added dockerfiles and pip dependencies for:

1. Speculative decoding workload for efficient LLM inference.
2. Model post-training model compression workload (quantization for now) for LLM deployment.