Integrate lm-eval into GuideLLM

[rgreenberg1]

Description
We've talked about adding in lm-eval to become a native pathway in GuideLLM to enable developers to run both benchmarks and accuracy evaluations in just one tool. This ticket will outline the necessary work to integrate lm-eval into GuideLLM.

Acceptance Criteria

• Package up the lm-eval-harness library into GuideLLM
• Create new command: guidellm-eval that can then take the same input parameters as lm-eval:
• • model hf \
• • model_args pretrained=EleutherAI/gpt-j-6B \
• • tasks hellaswag \
• • device cuda:0 \
• • batch_size 8

[markurtz]

@jennyyangyi-magic to scope this out a bit more and update