Enabled where for bool tensor on CUDA (#26430)

Summary:
Enabled "where_cuda" for bool tensors on CUDA
Fixing pytorch/pytorch#26247
Tested via unit tests
Pull Request resolved: pytorch/pytorch#26430

Differential Revision: D17464181

Pulled By: izdeby

fbshipit-source-id: cbb09925753b2e6f35e7400da3243d4d3fc86b69

aten/src/ATen/native/cuda/TensorCompare.cu

@@ -48,7 +48,7 @@ Tensor _s_where_cuda(
     const Tensor& self,
     const Tensor& other) {
   Tensor ret = at::empty(self.sizes(), self.options());
-  AT_DISPATCH_ALL_TYPES_AND(at::ScalarType::Half, ret.scalar_type(), "where_cuda", [&] {
+  AT_DISPATCH_ALL_TYPES_AND2(at::ScalarType::Half, at::ScalarType::Bool, ret.scalar_type(), "where_cuda", [&] {
     where_cuda<scalar_t>(ret, condition, self, other);
   });
   return ret;