docs: prominent input-format warning (this is a gene-ID model, not nucleotide)

fix: add [UNK] (and [PAD]) to WordLevel vocab to enable text-mode encoding

Add working list-of-tokens inference example (the WordLevel tokenizer has no pre_tokenizer; text-mode encode fails)

Add model card

Upload MolCrawl RNA BERT medium model (60k steps)

initial commit