[ESSnmx] Use more accurate domain types for sample

[jl-wynen]

There are two issues with how the sample rotation is encoded:

1. The workflow uses NXcrystal to identify the sample rotation. That class is meant for analysers (monochromators). So our code should be changed to use NXsample.

2. The sample rotation is stored as a Position. But that is not accurate, there should be a dedicated rotation / angle type. In spectroscopy we use this: https://github.com/scipp/essspectroscopy/blob/8676040cb9c693f8a5d7d84330fe5c0b87119db4/src/ess/spectroscopy/types.py#L114 We should move this into ESSreduce and optionally rename it if others prefer 'rotation' over 'angle'.

[YooSunYoung]

There is a rotation_angle field in the NXsample definition so we can just use them maybe.

There is also NXrotation.
Maybe we should use this for the output as well.

[jl-wynen]

For rotation_angle, the docs say

Optional rotation angle for the case when the powder diagram has been obtained through an omega-2theta scan like from a traditional single detector powder diffractometer. Note, it is recommended to use NXtransformations instead.

So I don't think this is what we want. You could use the NXtransformations of the sample to encode rotation. But that would be non-trivial to read.

There are also sample_orientation, orientation_matrix, and ub_matrix in NXsample. Are you encoding the angle of the goniometer or the actual crystal orientation that you get from fitting?

I don't know how and where NXrotations is used but it doesn't seem to be part of NXsample.

In any case, this is a question about how to store this data in a file. The answer depends on what DIALS can read. But this PR is about how to store that information in a workflow.