Adding `safetensors` variant of this model

docs: remove blockquote from experimental note

docs: add links to original repo and paper, add experimental proviso

feat: add LatinBertTokenizerFast with word_ids() support

chore: re-upload model weights (pytorch_model.bin)

chore: add HF model repo files (config, tokenizer, encoder, README)

refactor: extract shared case study utils and move data to tracked paths

fix: handle >512 token sentences and add MPS device support

test: add contextual nearest neighbors case study (Bamman & Burns §4.4)

feat: make benchmarks model-agnostic with --model-path option

chore: register pytest slow marker

test: add WSD case study reproduction (Bamman & Burns Table 2)

test: add POS tagging case study reproduction (Bamman & Burns Table 1)

test: add infilling case study reproduction (Bamman & Burns Table 3)

Fix tokenizer ID offset: reserve IDs 0-4 for BERT special tokens

Initial: HF-compatible Latin BERT tokenizer (Bamman & Burns 2020)