Fixing minor bugs in loading datasets

[yeomjy]

1. load_dataset(HF_DATASET[args.dataset], ...) -> load_dataset('json', data_files=HF_DATASET[args.dataset], ...) for each evaluation code
2. Current mbpp_intrinsic_eval.py loads hard dataset when not args.hard. Fixed this.
3. In mbpp_utdebug.py, unused code_src = load_dataset(...) variable exists and makes an error. Removed this.