Expand/update data/train_split.jsonl

Update train split

Fix: correct train split (stratified 90/10 from full 200, seed 42)

Move 90% split to train_split.jsonl (avoids clobbering train.jsonl)