CVE-2025-30402: ExecuTorch vulnerable to Heap-based Buffer Overflow attack

The vulnerability lies in the calculate_nbytes function in runtime/executor/method_meta.cpp. The commit 93b1a0c15f7eda49b2bc46b5b4c49557b4e9810f directly patches this function to add integer overflow checks. The vulnerability description mentions a heap-buffer-overflow during the loading of ExecuTorch methods, which aligns with the purpose of calculate_nbytes—to determine the memory size required for a tensor. The patch replaces the simple multiplication of dimensions with a version that checks for overflow at each step. The added test case TensorInfoSizeOverflow in runtime/executor/test/method_meta_test.cpp further confirms that the vulnerability is related to an overflow in size calculation by explicitly testing for it. The vulnerable function executorch::runtime::calculate_nbytes is called during the model loading process, and an integer overflow within it leads to the allocation of an incorrectly sized buffer, causing the heap overflow when tensor data is loaded.