Support uncertainty propagation for NDCube reprojection

[jmbhughes]

Describe the feature

During the SunPy Community Meeting on 2025-07-16, we discussed that NDCube's reproject function drops the uncertainty. @DanRyanIrish asked me to summarize here so we can continue the conversation. It would be good to handle uncertainty more robustly, but that requires caution and may require changes in reproject for full support. PUNCH specifically would benefit from propagating uncertainty during reprojection since we do many reprojections and utilize the uncertainty as a weighting factor when combining images.

Proposed solution

There are a couple options:

1. The more correct path is to have reproject handle uncertainties or at least supply the information about pixel areas and number of pixels combined for the uncertainty propagation. Then, ndcube could utilize that reproject feature.
2. The more hacky way is to simply reproject the uncertainty. It's not strictly correct but it's better than nothing and could be good enough in many uses cases.