support for torch 2.1

[gunsodo]

As it is mentioned in the wiki, the package might not work well with torch<2.2. One can reproduce the problem with torch=2.1 when torch.export has been first released, following the tutorial provided in the guide.

Is it possible to limit the torch version to >=2.2 in the pyproject.toml so that the error will be raised during the installation? Or alternatively, is it reasonable to have torch.export.ExportGraphSignature converter (from 2.1 to >=2.2)? Below is the trace:

Traceback (most recent call last):
  File ".../test_torch21_model_explorer.py", line 11, in <module>
    model_explorer.visualize_pytorch('mobilenet', exported_program=ep)
  File ".../lib/python3.11/site-packages/model_explorer/apis.py", line 82, in visualize_pytorch
    cur_config.add_model_from_pytorch(
  File ".../lib/python3.11/site-packages/model_explorer/config.py", line 84, in add_model_from_pytorch
    adapter = PytorchExportedProgramAdapterImpl(exported_program, settings)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File ".../lib/python3.11/site-packages/model_explorer/pytorch_exported_program_adater_impl.py", line 33, in __init__
    self.inputs_map = self.get_inputs_map()
                      ^^^^^^^^^^^^^^^^^^^^^
  File ".../lib/python3.11/site-packages/model_explorer/pytorch_exported_program_adater_impl.py", line 97, in get_inputs_map
    self.ep.graph_signature.input_specs, input_tensors
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'ExportGraphSignature' object has no attribute 'input_specs'

I am happy to open a PR for either approach, but I am not confident with the latter approach as I am not a Torch expert and there may be other issues following this one as well.

Anyway, thank you very much for creating this wonderful package!