fix: also use `add_prefix_space = True` in tokenizer config

feat: use RoBERTa tokenizer to (hopefully) fix some tokenization problems for token classification tasks

Set tokenizer "model_max_length" property to 8192 (#9)

undo last commit

Add `"add_prefix_space": true,`; this allows for much stronger token-level performance (e.g. NER, ColBERT) (#10)

Update tokenizer: Set lstrip=True for [MASK]

Upload ModernBERT-large model