NXP test_split_group_convolution__2d_1_group_3 failing on release branch CI

[GregoryComer]

🐛 Describe the bug

After pinning the pytorch and pytorch domain library dependencies on the release/1.0 branch, the NXP test_split_group_convolution__2d_1_group_3 unit test is failing with an assertion error on an np.allclose call (example failure).

We need to triage this issue to determine if this is a flaky test or legitimate regression.

======================================================================
FAIL: test_split_group_convolution__2d_1_group_3 (test_split_group_convolution.TestSplitGroupConvolution)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/parameterized/parameterized.py", line 620, in standalone_func
    return func(*(a + p.args), **p.kwargs, **kw)
  File "/pytorch/executorch/backends/nxp/tests/test_split_group_convolution.py", line 109, in test_split_group_convolution__2d
    assert np.allclose(out1, out2, atol=2.0e-7)
AssertionError

Versions

Failing on 473cb07 in the HUD (https://hud.pytorch.org/pytorch/executorch/commit/473cb079a5f0ad8941945ffbbfcc30403d546032).

cc @digantdesai @JakeStevens @robert-kalmar

[GregoryComer]

@robert-kalmar Could you take a look at the test failure and determine if there is a regression? Thanks!

[digantdesai]

Seems green on main but started failing on release/1.0 after #14642