How to retain BN after the conv?

In coreml export, the BN is fused into conv, is there a way to retain BN after the conv instead of fused into conv?
CoreML model:
ONNX model:

Start by reproducing the Core ML export described in the issue and compare its graph with the linked ONNX model. Determine whether the converter exposes an option to prevent batch-normalization fusion; done means documenting or implementing a supported way to retain the BN layer and verifying the resulting model structure.

Written by the indexing model from the issue text.