[Documentation] [questinon] How to trace intermediate tensor shapes: can logging level help here?

### Describe the documentation issue

For some inputs our model throws in onnxrt with some shape mismatch. So it seems that our tests don't cover all graph paths.

Is it possible to ask onnxrt to trace/print all intermediate tensor shapes (maybe via some InferenceSession options / env variable)? I looked up logging levels https://onnxruntime.ai/docs/api/c/group___global.html#ga1c0fbcf614dbd0e2c272ae1cc04c629c, but it's not written there what exactly they affect.

This probably would have worked, but our execution fails, so not sure if the intermediate outputs are still preserved in this case: https://github.com/microsoft/onnxruntime/issues/1455#issuecomment-1050615116

### Page / URL

_No response_

Start with the linked ONNX Runtime global logging documentation and the referenced issue comment. Determine whether logging levels or InferenceSession options expose intermediate tensor shapes when execution fails, and document the supported tracing method and the effect of each relevant logging level.

Written by the indexing model from the issue text.