Handle banks via workflow graph instead of via extra dimension

[SimonHeybrock]

The current McStas loader loads all banks and concats the events, https://github.com/scipp/essnmx/blob/c5a2520e1ac6b209c855552729c6b0076abf4453/src/ess/nmx/mcstas_loader.py#L50-L60 subsequently folding after grouping by pixel https://github.com/scipp/essnmx/blob/c5a2520e1ac6b209c855552729c6b0076abf4453/src/ess/nmx/mcstas_loader.py#L134

• Change this so only a single bank is handled in the code (loading, reduction steps).
• Multiple banks should be handled via Sciline, i.e., in the workflow graph. This will, e.g., allow for processing only a single bank at a time, saving memory.
  • We can use a param table to define the detector banks that should be processed.
• Saving to NeXus may need to change slightly. Need to check with the IDS.
  • I do not understand the current code, the panel dimension is first removed in https://github.com/scipp/essnmx/blob/c5a2520e1ac6b209c855552729c6b0076abf4453/src/ess/nmx/reduction.py#L245-L247 and then a panel dim of length 1 is added back into the dataset when saving, in https://github.com/scipp/essnmx/blob/c5a2520e1ac6b209c855552729c6b0076abf4453/src/ess/nmx/reduction.py#L151-L153.

[YooSunYoung]

Saving to NeXus may need to change slightly. Need to check with the IDS. I do not understand the current code, the panel dimension is first removed in

Yeah.. it's because what dials, the software that handles the next step, expects this as far as I understood...?