Fix fuse_pad_into_conv incorrectly fusing when Pad contains negative values

[Copilot]

The fuse_pad_into_conv optimization would fuse a Pad node into Conv even when pads contained negative values (valid for Pad as crop operations, but invalid for Conv's pads attribute), producing an invalid model.

Changes

• _fuse_pad_into_conv.py: Added a guard in _FuseConvPadBase.check() to bail out when any computed pad value is negative, preventing fusion that would produce an invalid Conv node:

  if any(p < 0 for p in self._pads_list):
      return check_result.fail(f"{pads.name} must not contain negative values.")

• _fuse_pad_into_conv_test.py: Added a test case to test_unsupported_fuse_pad_into_conv verifying that fusion is rejected when spatial pads are negative.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>fuse_pad_into_conv optimization incorrectly fuses when 'pads' contain negative values</issue_title>
<issue_description>converting the attached ONNX model with the following script:

from onnxscript import ir
from onnxscript.optimizer import optimize

ir_model = ir.load("input.onnx")
optimized_model = optimize(ir_model)
ir.save(optimized_model, "output.onnx")

results in the following output model:

The Conv layer has negative pad values, which is not allowed per spec. This optimization should check for negative pad values.

input.onnx.zip

versions of relevant packages:

ml_dtypes         0.5.4
onnx              1.20.1
onnx-ir           0.1.16
onnxscript        0.6.2
```</issue_description>

## Comments on the Issue (you are @copilot in this section)

<comments>
</comments>

• Fixes fuse_pad_into_conv optimization incorrectly fuses when 'pads' contain negative values #2838

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.