MXFP8 MoE training prototype graduation tracker

[danielvegamyhre]

• [P0] Support emulated mode Support emulated mode for mxfp8 moe training to support non-sm100 CI or dev env #3598
• [P0] Fix torch reference for per group blocked layout for groups along K
  • [mxfp8 moe training] fix torch ref impl of SF blocked layout per group along K #3603
• [P1] Add B200 to CI [BE] Add B200 runner to CI #2964 and add tests to workflow triggered by label
• [P0] Unify mxfp8 dense + moe code/configs
• [P0] Expose wgrad_with_hp recipe in quantize_ api and torchtitan
  • [mxfp8 moe training] export wgrad_with_hp recipe to quantize_ api #3611
• [P1] Expose use_triton_for_dim0_cast knob in quantize_ api and torchtitan
  • [mxfp8 training] add new configurable params now exposed by torchao torchtitan#2251
• [P1] Remove use_cuda_kernel_for_blocked_layout knob in quantize_ api, autoselect instead
  • [mxfp8 moe training] remove use_cuda_kernel_for_blocked_layout param #3735
• [P1] Ship gb200 compatible wheels
• [P0] Add better documentation for how to use each building block individually, padding group sizes, etc
  • [mxfp8 moe training][docs] add tutorial for training with MXFP8 expert parallel #3752