trunk/838676a9d695c331749ef7d5868f3a9224f7e93a: fix: compilation issues in mem_eff_attention with clang (#183102)

In file included from third_party/py/torch/aten/src/ATen/native/transformers/cuda/mem_eff_attention/kernel_forward.h:10:

In file included from third_party/py/torch/aten/src/ATen/cuda/PhiloxUtils.cuh:4:

third_party/py/torch/aten/src/ATen/cuda/detail/UnpackRaw.cuh:17:1: error: implicit instantiation of undefined template 'std::tuple<unsigned long, unsigned long>'

   17 | unpack(at::PhiloxCudaState arg) {

      | ^

third_party/crosstool/v18/stable/src/libcxx/include/__fwd/tuple.h:30:7: note: template is declared here

   30 | class tuple;

      |       ^

third_party/py/torch/aten/src/ATen/native/transformers/cuda/mem_eff_attention/kernel_forward.h:719:45: error: missing 'typename' prior to dependent type name 'MM1::LayoutB' (aka 'cutlass::layout::RowMajor')

  719 |             typename MM1::IteratorB::Params{MM1::LayoutB(p.v_strideM)},

      |                                             ^~~~~~~~~~~~

third_party/py/torch/aten/src/ATen/native/transformers/cuda/mem_eff_attention/kernel_forward.h:1013:45: error: missing 'typename' prior to dependent type name 'MM1::LayoutB' (aka 'cutlass::layout::RowMajor')

 1013 |             typename MM1::IteratorB::Params{MM1::LayoutB(p.v_strideM)},

      |                                             ^~~~~~~~~~~~

third_party/py/torch/aten/src/ATen/native/transformers/cuda/mem_eff_attention/kernel_backward.h:2404:5: error: reference to overloaded function could not be resolved; did you mean to call it?

 2404 |     MatmulQK::Mma::prologue(

      |     ^~~~~~~~~~~~~~~~~~~~~~~