lora unloading initial

Improved precision / reduced frequency of nan outputs, allow bf16 t5, f32 rmsnorm, larger clamp

Remove f8 flux, instead configure at load, improved quality & corrected configs

Dynamic swap with cublas linear / optional improved precision with vram drawback

Remove unnecessary code, hide prints behind debug flag, hide warnings

Fix non-prequantized inference

Fix non-offload inference & add option to load from prequantized flux

Add offloading & improved fp8 inference.

Updates..

Add img2img, more options, gradio interface

initial commit