Fix noncontiguous AOT inputs

[jloftin-nv]

Description

Input.from_tensor rejects non-contiguous / non-channels-last tensors unless disable_memory_format_check=True. The torch.compile backend sets that flag; every dynamo.compile AOT path does not, so the same inputs succeed on one entry point and fail on the other.

Pass disable_memory_format_check=True through the AOT prepare_inputs sites to match the backend behavior.

Type of change

Please delete options that are not relevant and/or add your own.

• Bug fix (non-breaking change which fixes an issue)

Checklist:

• [X ] My code follows the style guidelines of this project (You can use the linters)
• [X ] I have performed a self-review of my own code
• [X ] I have commented my code, particularly in hard-to-understand areas and hacks
• [ X] I have made corresponding changes to the documentation
• [ X] I have added tests to verify my fix or my feature
• [ X] New and existing unit tests pass locally with my changes
• [ X] I have added the relevant labels to my PR in so that relevant reviewers are notified

[lanluo-nvidia]

• P1 — The top-level Dynamo AOT entry points still reject non-contiguous tensors before reaching these updated functions. py/torch_tensorrt/_compile.py:340,
  py/torch_tensorrt/_compile.py:438, and py/torch_tensorrt/_compile.py:564 call prepare_inputs(...) without disable_memory_format_check=True. Thus
  torch_tensorrt.compile(..., ir="dynamo"), top-level cross-compilation, and convert_method_to_trt_engine still raise ValueError for the same transposed
  inputs, so the change only fixes direct torch_tensorrt.dynamo.* APIs. The test covers only the direct compile path.

[jloftin-nv]

• P1 — The top-level Dynamo AOT entry points still reject non-contiguous tensors before reaching these updated functions. py/torch_tensorrt/_compile.py:340,
  py/torch_tensorrt/_compile.py:438, and py/torch_tensorrt/_compile.py:564 call prepare_inputs(...) without disable_memory_format_check=True. Thus
  torch_tensorrt.compile(..., ir="dynamo"), top-level cross-compilation, and convert_method_to_trt_engine still raise ValueError for the same transposed
  inputs, so the change only fixes direct torch_tensorrt.dynamo.* APIs. The test covers only the direct compile path.

Fixed the other pathways