Support power of 2 scaling factors in float8 training and use e4m3 everywhere #7884

Job	Run time
test-nightly (CPU Nightly, linux.4xlarge, --pre torch==2.7.0.dev20250122 --index-url https://down... / linux-job	10m 53s
test-nightly (CUDA Nightly, linux.g5.12xlarge.nvidia.gpu, --pre torch==2.7.0.dev20250122 --index-... / linux-job	47m 35s
test (CPU 2.4, linux.4xlarge, torch==2.4.0 --index-url https://download.pytorch.org/whl/cpu, cpu) / linux-job	9m 13s
test (CPU 2.5.1, linux.4xlarge, torch==2.5.1 --index-url https://download.pytorch.org/whl/cpu, cpu) / linux-job	11m 23s
test (CPU 2.3, linux.4xlarge, torch==2.3.0 --index-url https://download.pytorch.org/whl/cpu, cpu) / linux-job	7m 34s
test (CUDA 2.3, linux.g5.12xlarge.nvidia.gpu, torch==2.3.0, cuda, 12.1) / linux-job	31m 36s
test (CUDA 2.4, linux.g5.12xlarge.nvidia.gpu, torch==2.4.0, cuda, 12.1) / linux-job	47m 9s
test (CUDA 2.5.1, linux.g5.12xlarge.nvidia.gpu, torch==2.5.1 --index-url https://download.pytorch... / linux-job	1h 13m 22s
	3h 58m 45s

Provide feedback