DynamicToFixedShape: use ort shape infer tool, accept 0 as dim_value

[jambayk]

Describe your changes

• onnxruntime.tools.onnx_model_utils.fix_output_shapes cannot handle large models (Unable to export model by using --device npu --provider QNNExecutionProvider #1595), so we use the ort shape infer helper and handle the logic ourselves. This also means it can now handle models with contrib operators too.
• allow passing 0 as dim_value. this case is possible when creating a prompt processing model from a dynamic shaped llm where we want to make the past kv cache empty.

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