Update on "Use safe_numel() in et"

The wasm bindings' assert_valid_numel() helper previously computed numel without overflow detection, letting a malicious tensor shape wrap to a small value that bypassed the data-size check. Replace with safe_numel() (returns Result<ssize_t>) and propagate the error via THROW_IF_ERROR as a JavaScript exception, matching the rest of the wasm error plumbing.

Authored with Claude.

Differential Revision: [D102082911](https://our.internmc.facebook.com/intern/diff/D102082911/)

[ghstack-poisoned]

Author: Github Executorch

runtime/core/exec_aten/exec_aten.h

@@ -114,7 +114,7 @@ inline ssize_t compute_numel(const SizesType* sizes, ssize_t dim) {
       c10::multiply_integers(c10::ArrayRef<SizesType>(sizes, dim)));
 }
 
-ET_EXPERIMENTAL inline ::executorch::runtime::Result<ssize_t> safe_numel(
+inline ::executorch::runtime::Result<ssize_t> safe_numel(
     const SizesType* sizes,
     ssize_t dim) {
   ET_CHECK_OR_RETURN_ERROR(

runtime/core/portable_type/tensor_impl.h

@@ -296,12 +296,12 @@ ssize_t compute_numel(
     ssize_t dim);
 
 /**
- * EXPERIMENTAL. Compute the number of elements based on the sizes of a tensor.
+ * Compute the number of elements based on the sizes of a tensor.
  * Returns Error::InvalidArgument if any intermediate multiplication would
  * overflow ssize_t, or if a size is negative. Prefer this over compute_numel()
  * for paths that can propagate an Error upward.
  */
-ET_EXPERIMENTAL ::executorch::runtime::Result<ssize_t> safe_numel(
+::executorch::runtime::Result<ssize_t> safe_numel(
     const ::executorch::runtime::etensor::TensorImpl::SizesType* sizes,
     ssize_t dim);