Bit depth image

[phyy-nx]

New PR superseding #1262 with additional edits from code camp review

From @graeme-winter:

Fixes #1258 by including it, using effectively the same definition as is currently in the field from DECTRIS

[yayahjb]

Worth doing

[tacaswell]

The name is over-determined by DECTRIS already writing with this name so we have to accept it as is.

I am still confused by why this not redundant with asking the dataset what type it has.

Some of the linked discussions make the good point that there is not a way to note if the data is signed or unsigned. Might be notionally better to have image_datatype, but as noted above we are being jammed by DECTRIS so we should just accept this and move on.

[phyy-nx]

Proposal is instead to try bit_depth_STORED in the base class, but this convention requires #1436

[benajamin]

I don't like the proposal to to add bit_depth_STORED. It is implementing a feature whose only current use case is to make the name flexible enough to legitimise the "bit_depth_image" that DECTRIS is already writing. We don't need to make it ambiguous with this feature. We should just give the field a simple name (e.g. "bit_depth_stored") and and if that name is not "bit_depth_image", then we should both define and deprecate "bit_depth_image".

[phyy-nx]

How about this? Use bit_depth_stored for NXdetector and bit_depth_image for NXmx, and just accept a redundant field in NXmx. Awkward, but it's less awkward than bit_depth_STORED for sure.

[phyy-nx]

Chatting more with @benajamin, go with the deprecated bit_depth_image instead.

[tacaswell]

I think we should either bless exactly what dectris/MX is doing in the most minimal way or consider ways to more accurately specify the data types (signed int vs unsigned int vs float (and if you want to open up floats there are non-standard floats (that you can write to hdf5) that use varying fractions of available bits for the exponent and mantissa and rgb (is a color image 8 bits or 24 bits under bit_depth_image?) of both what was collected from the hardware and what is written to the disk.

The first is pragmatic but wrong, the second is a lot of work but likely the correct thing to do.

[phyy-nx]

Would changing to NX_NUMBER suffice for all the edge cases? Could do that for both bit_depth_image and bit_depth_stored.

[tacaswell]

still approve after recent changes.