Add SWE-Bench Verified-466 (no matplotlib) eval set

Add swe_gym_train_eval_100.jsonl (3rd eval set, 100-instance held-out from train)

drop redundant full swe_gym.jsonl; training uses pre-filtered swe_gym_train_minus_val100.jsonl

Upload swe_gym.jsonl with huggingface_hub

Upload swe_gym_train_minus_val100.jsonl with huggingface_hub

Upload swe_gym_val_100.jsonl with huggingface_hub

Upload swe_bench_django.jsonl with huggingface_hub

initial commit