Disable CPU offload when running on Hugging Face ZeroGPU to prevent device mismatch errors

Implement CPU and GPU offloading for plugins and VRAM manager

Fix device mismatch error in FLUX pipelines by calling .to(device)

Integrate ByteDance Hyper-SD 8-step LoRA for 3.5x faster FLUX Dev image generation

Fix quantization config loading for FLUX and FLUX Fill pipelines

Fix directory nesting and commit updated plugin code

Add modular NanoBanana Engine and plugins for Kaggle GPU hosting