[Feature Request] How to directly define a model that includes a qgemm node？

[R-Ceph]

Describe the feature request

In this issue (#19503), we found that sometimes it's not possible to directly convert gemm to qgemm.

Moreover, on this page (https://onnx.ai/onnx/operators/onnx__MatMulInteger.html) I found that QGemm is a custom operator of onnxruntime, which means it's also not possible to directly create a QGemm node using onnx.helper.make_node.

Therefore, I would like to consult on how to directly define a model that includes a qgemm node, rather than obtaining it through conversion?

Describe scenario use case

Sometimes it's not possible to directly convert gemm to QGemm, and I want to try QGemm instead of QLinearMatMul.