add check to ensure CLIP is embedded before saving to prevent crash during LoRA extraction

[GlenCarpenter]

#1330

This provides a safeguard in the Python code to ensure base_model_clip and other_model_clip are set before extracting the embedded CLIP.
Flux Checkpoints in the Stable-Diffusion directory are treated as regular checkpoints, and the check for doClip is based on the directory structure:

// in ComfyUIWebAPI.cs
bool doClip = !baseModelData.IsDiffusionModelsFormat && !otherModelData.IsDiffusionModelsFormat

...
...

// in T2IModel.cs
public bool IsDiffusionModelsFormat
    {
        get
        {
            string cleaned = OriginatingFolderPath.Replace('\\', '/').TrimEnd('/').ToLowerFast();
            return cleaned.EndsWithFast("/unet") || cleaned.EndsWithFast("/diffusion_models"); // Hacky but it works for now
        }
    }

It seems like a long term goal would be to check whether or not checkpoints in the Stable-Diffusion directory contain an embedded text encoder, and what that text encoder is (like CLIP or T5). As far as I know that is not reliably present in checkpoint metadata.