Using SD 3.5 Large Fine Tuned Models with Embedded Clip L + Clip G but with External T5

### Feature Idea

Currently it either loads all Clip L + Clip G + T5 from single safetensors file or

It loads all 3 externally

However what I want is loading Clip L + Clip G from trained safetensors file and loading T5 from external file

Because training T5 is usually useless. But training Clip models helps

Kohya supporting all these options but I talked with SwarmUI developer and checked from ComfyUI back-end, ComfyUI doesn't support such separate loading

You can see Kohya latest pull request here : https://github.com/kohya-ss/sd-scripts/pull/1960/files#diff-6b605dc86514575eab565c7ba50409469c1b6d4e88013e38ece90d9735340125

It works perfect I tested genearting a combined safetensors file with

DiT + Clip L + Clip G - ComfyUI can't use T5 in this case
and
DiT + Clip L + Clip G + T5 - ComfyUI uses T5 in this case but it makes files even larger totally unnecessarily

Thank you