Fix Llama 3.1 finetuning example

[alex000kim]

Simplifies Llama 3.1 LoRA finetuning by generating platform-specific configs dynamically using tune cp instead of maintaining static config files.
Currently, using static configs breaks this example on certain platforms/GPUs.

Changes

• Remove configs/ folder and static YAML files
• Generate platform-specific configs at runtime with tune cp llama3_1/${MODEL_SIZE}_lora
• Update lora.yaml to use uv for dependencies