[dynamo exporter] Support string in dynamic_shapes

[titaiwangms]

Describe your changes

With pytorch/pytorch#146321, torch.onnx.export(.., dynamo=True) now can support string in dynamic_shapes, which fits better with Olive driven with configuration.

Major changes:

• Add support for string in dynamic_shapes
  • Move dynamic_shapes pre-process to io_config.py (like dynamic_axes)
  • Get rid of the post-process of making [str, int, int] -> torch.export.Dim(str, max=int, min=int). torch.onn.export(..., dynamo=True) can now take string.
  • Leverage Optimum to auto-generate dynamic_shapes when Optimum models is requested.
  • KV cache support

Pitfall:

• When dynamic_shapes targets kwargs, both of them need to follow the order of model.forward signature. onnx/conversion.py provides naive approach to sort them, but users should be aware of this.

Minor changes:

• Move onnxscript (released) to the official requirement.txt
• dynamic_shapes with string is supported since torch 2.7

Checklist before requesting a review

• Add unit tests for this change.
• Make sure all tests can pass.
• Update documents if necessary.
• Lint and apply fixes to your code by running lintrunner -a
• Is this a user-facing change? If yes, give a description of this change to be included in the release notes.
• Is this PR including examples changes? If yes, please remember to update example documentation in a follow-up PR.

(Optional) Issue link