[LLM Bench] Additional Config for PT Benchmarking in LLM Bench

[anzr299]

I would like to request the ability to pass additional configuration for pytorch framework in LLM bench when loading a pipeling.
While loading SD3 pipeling without text_encoder_3 and tokenizer_3, it is expected to pass these values as None while loading SD3 pipeline.
At the moment loading such a pipeline requires modifying This line like so:
pipe = model_class.from_pretrained(model_path, text_encoder_3=None, tokenizer_3=None)