Add diffusion FLUX fp8_static quantization

[yghstill]

Features

• Support FLUX's Transformer block fp8 static quantization calibration.
• Export quantization model as safetensors format.
• Support FLUX's quantization inference, load quant config from angelslim_config.json, for example:

from angelslim.engine import InferEngine
slim_engine = InferEngine()
slim_engine.from_pretrained(model_path="youu/quant/angelslim/model/")
output = slim_engine.generate("A beautiful landscape with mountains and a river.")

TODO

• For better reviewability, the combined compression strategy (e.g., cache + quant) will be submitted in the next PR.