Fix stack buffer overflow when XNNPACK tensor has too many dims (#3233)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/3233

Noticed this overflow when I was looking through the XNNPACK backend.

I am not very familiar with executorch or XNNPACK, so please be critical in review! In particular, my test is crashing in xnn_delete_runtime and I don't know why; it was previously getting a UBSAN failure for the stack buffer overflow before I patched XNNExecutor.cpp

Reviewed By: digantdesai, mcr229

Differential Revision: D56450593

fbshipit-source-id: d0f2b4d5dc44e08f9679560d4a6adde7f1559173