Scheduling uses ComfyUI's hook system, which crashes at sample time on
quantized Linear layers ('Linear' object has no attribute 'temp'/'weight_scale').
Add quant_utils.is_quantized_model() and, when a schedule is requested on a
GGUF/fp8 model, fall back to a flat-strength apply with a clear warning in the
node output instead of crashing. Covers the V2 combined Selective Loaders and
the standalone ScheduledLoRALoader. Addresses #26, #41, #51, #33.
NOT FOR RELEASE YET - detection heuristics need testing on real GGUF/fp8 models.
