Load NXdata with auxiliary signals with different dims

[jl-wynen]

NXdata can contain auxiliary signals in addition to the regular signal. When they are present, we load the whole NXdata group as a Dataset. But the loader requires that all signals and coords share dimensions with the main signal. This is not the case for monitors in ESS files. They now contain a frame_total aux signal that has a different dimension.

There are a couple of options for loading this:

• Extend the loader to build a dataset with the proper dims
  • Preserves current interface.
  • Potentially tricky to implement.
  • Wastes loading the aux signal if unused in a workflow.
• Always quietly fall back to loading as a data group
  • This requires updates to the generic nexus workflow to extract and assemble the signal and aux signals.
  • Simplest fix.
  • Wastes loading the aux signal if unused in a workflow.
• Drop aux signals.
  • This means that group[()] loads the familiar data array.
  • It avoids loading data that will be discarded immediately after in reduction workflows.
  • I don't think we use aux signals anywhere, so this wouldn't actually break anything.
  • But it becomes more difficult to access aux signals in the future.
  • Impl: Detect signal and its coords; drop everything else.

[jl-wynen]

We already load nxlogs as data groups of multiple data arrays. So we could do that here, too. But for NXlogs, the problem is that all sublogs have a time dim and those dims have different lengths. Here, we have shared dims and coords that can all be combined into a dataset.

[jl-wynen]

We can use an application definition to filter out aux signals, instead of hard-coding that in scippnexus. Then we can define that filter in essreduce.

[SimonHeybrock]

We already load nxlogs as data groups of multiple data arrays. So we could do that here, too. But for NXlogs, the problem is that all sublogs have a time dim and those dims have different lengths. Here, we have shared dims and coords that can all be combined into a dataset.

scipp.Dataset does not support this anymore, see https://scipp.github.io/development/adr/0017-restrict-dataset-to-items-with-matching-dimensionality.html. If this solution is what you chose, I would recommend to use scipp.DataGroup, coords can share memory via underlying variables.