Disable qdq to mnb fusion in test_mnb_to_qdq

[jambayk]

Describe your changes

Latest ORT has QDQ to MatMulNBits rules for more cases now. We want to disable this fusion in the test_mnb_to_qdq test since we are trying to compare the original MNB model with the replacement QDQ model.

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.

(Optional) Issue link

[Copilot]

Pull request overview

Updates an ONNX pass unit test to prevent newer ONNX Runtime versions from fusing the replacement QDQ graph back into MatMulNBits, so the test can continue validating behavior of the QDQ replacement model.

Changes:

• Disables a specific ORT optimizer when creating the QDQ model InferenceSession in test_mnb_to_qdq.
• Adds a short comment explaining why the optimizer is disabled.