feat: add lorentz_correction to powder workflow

[jokasimr]

Adding the correction to the workflow graph was more messy than I expected.

Workflow/graph related issues

The guidelines say that Lorentz correction should be applied to CorrectedDetector.

But at that step in the workflow the events don't have dspacing (and Lorentz correction is a function of dspacing).
I don't know if this is a bug in EssDiffraction or if it is a bug in the guidelines.
I worked around the issue by removing CorrectedDetector, going directly from WavelengthDetector to DspacingDetector.

However, this makes the provider add_coords_masks_and_apply_corrections do a lot of things at once.
And it takes a lot of arguments that are not really related, so it's not very useful for understanding what the provider does.

The other option would be to add more domain types where each step is done separately. But that would add a lot of unnecessary domain types to the graph.

There's a tension between making the graphs show relevant intermediate results, and making the graphs explain what is being done to the data.

Issues in apply_lorentz_correction

The existing apply_lorentz_correction function could not be applied to data with variances.
For now I just replaced it with something that works with variances.

[jl-wynen]

I there a reason why we can't compute a dspacing coord earlier and assign it to WavelengthDetector? Then we could merge WavelengthDetector and DspacingDetector and apply the corrections afterwards. (Note that we are not binning in any of these steps anyway.)

[jokasimr]

I there a reason why we can't compute a dspacing coord earlier and assign it to WavelengthDetector? Then we could merge WavelengthDetector and DspacingDetector and apply the corrections afterwards. (Note that we are not binning in any of these steps anyway.)

Yeah that's also a possibility. It has the same drawbacks as the other version, and it makes the name WavelengthDetector a bit strange. But I agree it's a nicer split overall!

Another option is to also remove WavelengthDetector and do everything in one step:
Compute the coords, add the masks, and do the lorentz correction.

[jokasimr]

Have to keep tof coord because it is used by masking in later step.

I don't really see any reason to drop event_time_offset here except for performance.
Maybe that's reason enough.
On the other hand, that makes users unable to investigate correlations between event_time_offset and other coordinate if they want, which might be annoying.

[jl-wynen]

The types don't make sense. If the function was used as a provider like this, it would receive data that already contains d-spacing. And it the return type would clash with or bypass corrections. So I think you should just use sc.DataArray here.

[jl-wynen]

How can this be a dataset?

[jokasimr]

I don't know. The type checker told me it could be.

[jl-wynen]

It's definitely wrong. The input is a data array and all definitions of out are also data arrays.