Fix stale tech_id comment (only 0/1/2 valid after dropping tetrode slot)

examples: rewrite extract_embeddings.py to use from_pretrained() by default; mirror of HIPPIE/examples/

docs: link end-to-end script; sync dataset name to lisberger (single-s)

3-class technology vocabulary; remove tetrode reserved-slot note

Document released checkpoint as 3-class technology model (drop unused tetrode slot from public docs)

Fix Lisberger tech (16-ch Plexon s-Probe), remove false cross-species cerebellar-vocabulary claim, remove fabricated biorxiv DOI, clarify that released ckpt disables super-region/layer conditioning, paper-canonical dataset names and sample sizes

Correct dataset table (cell-type counts per paper); remove fabricated biorxiv DOI; canonical dataset names (Watson/Calvigioni/Ramachandran/Lakunina/Yu); Neuropixels 1.0 attribution

Expand training-data table to all 11 datasets; point install/usage at HIPPIE release

Use hippie.inference.HIPPIEClassifier + hippie.checkpoint API

Update README to use HIPPIEClassifier release API

Add umap_by_dataset.png

Add umap_coords.npz

Add embeddings.npz

Fix NaN ISI/ACG: nan_to_num before encoding

Add pretrained technology-conditioned checkpoint v1

Add embedding extraction utility

Add model card

initial commit