Reuse CLIP-loaded LLM for prompt assistance and in-workflow chat.

[Well-Made]

Since using LLMs for text encoding is becoming more common, I’d like to suggest adding the ability to reuse the LLM already loaded via the CLIP Loader. This would let users:

1. automatically enhance prompts (similar to how it’s done in LTX-2), and
2. chat with the model directly inside the workflow to get tips on prompt writing.
   Right now, this isn’t possible - existing third-party LLM nodes always require loading a separate model from a dedicated folder, even if a compatible LLM is already loaded.