Unloading Image Generation Models

Before calling the text generation API, the image generation model is unloaded from memory. In other words, this feature is designed to free up memory for the local LLM. This is currently implemented for A1111 only and has been verified to work specifically with forge neo.

Regarding the concern of text generation and image generation competing for memory: please handle the LLM side by configuring it to unload after a standby period of, for example, 1 second.