Fixed rslora scaling in lora_manager

[TheCodeWrangler]

Addressing issue mentioned in https://github.com/NVIDIA/TensorRT-LLM/issues/1668

When weights were trained using rslora scaling they should be scaled differently. Code initially was always normalizing by rank regardless of "use_rslora" flag in huggingface adapter_config.json file.

Scaling has also been updated in examples/hf_lora_convert.py

[byshiue]

Could you share a model trained by rslora?

[nv-guomingz]

Hi @TheCodeWrangler , thanks for your contributing. We've merged your contribution into code base and will add you into contributor list.