Qualcomm AI Engine Direct - Adding QNN backend support for diagonal core ATen op

[qti-horodnic]

Summary

Added support for the core ATen op diagonal using a decomposition pass and the permute, view, arange, index_select ops by:

1. Permute input so dim1 and dim2 are the last two dimensions.
2. Reshape (view) to flatten the last two dims: [..., M*N].
3. Compute flat diagonal indices via arange(start, end, stride).
4. Use index_select on the last dim with the computed indices.

Also made a small update to the new_op_development skill.

Test plan

python backends/qualcomm/tests/test_qnn_delegate.py -k TestQNNQuantizedOperator.test_qnn_backend_diagonal --soc_model SM8750 --host aisw-vm15-labsd --device 545ee4aa --build_folder build-android

python backends/qualcomm/tests/test_qnn_delegate.py -k TestQNNFloatingPointOperator.test_qnn_backend_diagonal --soc_model SM8750 --host aisw-vm15-labsd --device 545ee4aa --build_folder build-android

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20596

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit 58a639a with merge base 2af5638 ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[qti-horodnic]

@pytorchbot label "release notes: qualcomm"