add separate quantization primitives for float8 #5067

Job	Run time
test (H100, linux.aws.h100, --pre torch torchvision torchaudio --index-url https://download.pytor... / linux-job	23m 31s
test (SM-89, linux.g6.4xlarge.experimental.nvidia.gpu, --pre torch==2.7.0.dev20250122 --index-url... / linux-job	12m 1s
	35m 32s

Provide feedback