[Bug Fix] Improve binary search handling in JIdxForJOffsets

[iYuqinL]

Refactor binary search logic to handle cases with duplicate offsets correctly.

The original jidx from joffsets binary search can not handle the case that when joffsets has same values.
For example, joffsets is [0, 10, 10, 40], when idx = 10, the original code will wrongly set jidx to -1.

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: iYuqinL / name: Yuqin Liang (5c76d50, 6508f0d, 6fb9322, 77ee048)

[fwilliams]

Thanks a lot for the fix! I added a few comments about conforming to codestyle. You can see all the style fixes here: https://github.com/openvdb/fvdb-core/actions/runs/19194298081/job/55129073233?pr=325

If possible, could you also add a unit test in Python? You can add it to tests/unit/test_jagged_tensor.py

[iYuqinL]

@fwilliams @harrism hello, I have formatted the code using clang-format as you suggested.
And I have also added a unit test function to tests/unit/test_jagged_tensor.py, which named test_jidxforjoffsets.

[harrism]

Thanks so much for the great contribution! Can you please add a C++ test? Should only need a couple lines in an existing test.

[iYuqinL]

@harrism hello, I have add two unit tests to src/tests/JaggedTensorTest.cpp file, named JIdxFromJOffsetsCPU and JIdxFromJOffsetsCUDA. They test the jidx_from_joffsets function on CPU device and CUDA device.

I also run the unit test in my local env, without the kernel modification, the JIdxFromJOffsetsCUDA runs fail as expected.

with the kernel modification, both JIdxFromJOffsetsCPU and JIdxFromJOffsetsCUDA can run successfully.

[harrism]

Thank you for this excellent PR! We really appreciate your contribution @iYuqinL.

[iYuqinL]

@harrism hello, I discovered that my previous commits weren't GPG signed.
I re-push the signed commits, but it seems you need to review them again to get them through the merge process.

[harrism]

Thank you for this fix and the new test. We look forward to your future contributions!