Replace raw extraction with sanitized vLLM-ready truthfulness LoRA (pure lora_A/lora_B, modules_to_save stripped; source: HiTZ/Llama-3.1-8B-Instruct-multi-truth-judge)

Update README for sanitized vLLM-ready truthfulness adapter

Add truthfulness LoRA adapter extracted from HiTZ/Llama-3.1-8B-Instruct-multi-truth-judge (base: meta-llama/Llama-3.1-8B)

initial commit