fix(fuse_batchnorm): support convtranpose + bn fusion with group != 1

[AyoubMDL]

Fixes #2867

[justinchuby]

Thanks!

[Copilot]

Pull request overview

Fixes BatchNormalization fusion for grouped ConvTranspose so that optimization no longer crashes (broadcast/reshape mismatch) when group != 1, addressing #2867.

Changes:

• Implement group-aware weight scaling for ConvTranspose + BatchNormalization fusion.
• Add a guard to skip fusion for semantically invalid grouped ConvTranspose (when in_channels % group != 0).
• Expand tests to cover grouped Conv/ConvTranspose fusion and the invalid-model skip case.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
onnxscript/rewriter/rules/common/_fuse_batchnorm.py	Refactors weight scaling and adds correct grouped ConvTranspose handling + validation to avoid reshape/broadcast failures.
onnxscript/rewriter/rules/common/_fuse_batchnorm_test.py	Extends unit tests for grouped fusion behavior and adds a regression test for invalid grouped ConvTranspose.

[codecov]

Codecov Report

❌ Patch coverage is 91.89189% with 3 lines in your changes missing coverage. Please review.
✅ Project coverage is 71.97%. Comparing base (1077da7) to head (3c55666).
⚠️ Report is 11 commits behind head on main.

Files with missing lines	Patch %	Lines
...nnxscript/rewriter/rules/common/_fuse_batchnorm.py	87.50%	2 Missing and 1 partial ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2879   +/-   ##
=======================================
  Coverage   71.96%   71.97%           
=======================================
  Files         239      239           
  Lines       29224    29251   +27     
  Branches     2878     2880    +2     
=======================================
+ Hits        21031    21053   +22     
- Misses       7216     7219    +3     
- Partials      977      979    +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.