Added native GLiNER entity extraction to the MLX runner through `/api/extract` and the Go client, returning entity labels, scores, and Unicode offsets.
This includes tokenization, DeBERTa encoding, span scoring, model import and metadata, an export script, and documentation. Initial support targets float32 DeBERTa GLiNER `markerV0` models.
The build and affected tests pass. I verified Python checkpoint parity and HTTP extraction, including concurrent requests and input validation, on MLX CPU.
Bruce MacDonald
9594cb78ad