v0.2.0

ComfyUI support: a LoRA that cannot reach the model it is applied to now stops the workflow before sampling, with the reason.

ComfyUI skips every LoRA module with no counterpart in the loaded model (one console line each) and finishes the run as a success, so a LoRA made for another base model silently does nothing. Measured on a real ComfyUI 0.34.1: an Anima LoRA in an SDXL workflow logged 840 such lines and changed the image by 0.8/255 (the matching LoRA: 35.2).

With ENTAIL=load, entail counts how many LoRA modules reach the model and the text encoder using ComfyUI's own key map:

• none: stops with what the LoRA declares it was trained for and which model it met
• some: one line, the run goes on

Measured: both wrong directions (Anima <-> SDXL) stop; 22 right pairings pass with no false alarm; images are identical with entail on and off.

pip install -U entail-ai

Also: entail doctor recognises ComfyUI when run from its folder.