feat: retarget fine-tune pipeline to MiniCPM4.1-8B

feat: grounded, varied training explanations for MCQ and evaluation

fix: custom collator so EOS is learned (pad==eos masked stop tokens in v2)

fix: multi-task fine-tuning — single-task training broke MCQ format

feat: llama.cpp runtime ready for the Space (Llama Champion)

fix: verify Hub write access before training, fallback to user namespace

feat: Well-Tuned + Llama Champion groundwork