Potential inconsistency between `nvdiffrast` and this project in `interpolate` method

Hello! Thank you for your excellent work! We are currently migrating the FoundationPose algorithm to a non-ROS environment and noticed a potential inconsistency worth verifying.

In the original nvdiffrast library, the interpolate method determines p.instance_mode based on the ​shape dimensions of the input attr code reference. However, in the current migrated interpolate wrapper code reference, this check is implemented using the ​vector length of attr, which results in behavioral differences compared to the original implementation.

We observed this issue specifically during our implementation of the original algorithm's use_light rendering process, where it affects the interpolation of the diffuse_intensity_map. However, this discrepancy does not appear to impact the ​correctness of the current project’s outputs.

We would greatly appreciate your insights on this observation. If this is indeed an unintended discrepancy, we would be happy to submit a pull request to align the logic with the original behavior, pending your guidance.

Thank you for your time and consideration!