Add additional safeguards when loading sparse tensor

[TBrenetNV]

Describe the issue

When loading a model that uses sparse tensor but with incompatible dense and indices tensor shape, onnxruntime will crash.
Additional checks should be added when loading sparse tensors to ensure data can actually be loaded and avoid reading array pass their end. Having those safeguards will also help providing better error feedback.

onnxruntime/onnxruntime/core/framework/tensorprotoutils.cc

Line 1262 in c20ced4

for (size_t i = 0; i < n_sparse_elements; ++i) {

There is no check that cur_index is at least of size n_sparse_elements * rank

To reproduce

Create an inference session with the attached ONNX model, a crash will occur.

import onnxruntime
sess = onnxruntime.InferenceSession('wrong_sparse_shape.onnx', providers=['CPUExecutionProvider'])
# Crash

wrong_sparse_shape.zip

Urgency

No response

Platform

Windows

OS Version

10

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.16.1

ONNX Runtime API

Python

Architecture

X64

Execution Provider

Default CPU

Execution Provider Library Version

No response