Expose more params in main_export for customized ONNX export

[hann-wang]

What does this PR do?

We aim to provide more flexible control over the ONNX exporter, which includes:

• support for the ROCMExecutionProvider
• registration of custom operators
• the ability to use a customized export function besides torch.onnx.export

Currently, there are some challenges in achieving these objectives:

• providers and session_options are initialized in the fix_dynamic_axes and the _run_validation functions, which offers no interface for registering custom operators or specifying ROCMExecutionProvider.
• main_export loads models by names or paths, which is unfeasible to export a tailored model.
• disable_dynamic_axes_fix is not exposed in main_export. If custom operators are not registered in ONNXRuntime, fix_dynamic_axes will fail. To write ONNX models to output, we must be able to disable this function.
• Many 3rdparty model optimization tools have customized export functions other than torch.onnx.export, making a customizable export function for main_export more desirable.

This PR addresses the aforementioned challenges by:

• allowing a model object as input to main_export
• adding disable_dynamic_axes_fix, providers, session_options, and custom_export_fn params to main_export.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

Who can review?

@fxmarty, @echarlaix, @JingyaHuang, @michaelbenayoun