feat: enable fine-tuned model loading from Hugging Face Hub

fix: use base model for demo (remove LoRA adapter dependency)

feat: Multi-tool selection and robustness testing