Update libCZI and fix shape error

[toloudis]

We had an onprem image that bioio was badly failing to read.
Prior to this PR, bioio was reporting the shape as 5,60,180602,113778.
After updating libCZI, the shape was reported as 5,60,1248,1856. Much better! But the individual image planes were clearly sized at x=1848. It turned out that some of the channels had a different x offset than others (by 8 pixels), and the total x bounds spanned 1848+8 = 1856 pixels. Then when we go to load just one CZYX array from the data, there is a mismatch between the data shape and the expected total dask array shape. This would cause a crash inside bioio-czi.
So I also found and fixed the inconsistency here.

• Update libCZI (and pybind11 while we're at it) to pick up reader features for newer CZI files.
• some autoformatting
• remove short-cut way of getting the YX plane size for a scene, which was using boundingBox instead of the subblocks' logicalRects, which can result in inconsistent different results. The logicalrect method returns results that are consistent in shape with the data block returned when we read image data.

After this fix goes in, we have to do a release build of aicspylibczi, and then update bioio-czi to use it.

[toloudis]

this method which uses boundingBoxLayer0 can return a different result than the following code which uses the subblock logicalRects. Preferring the logicalRect gives consistency between expected shape and read shape. Therefore I am removing this short-cut attempt to read the scene YX shape

[sandraoluoch]

Hi Dan, I've reviewed the changes you've made and agree with them.

[SeanLeRoy]

🙌

[heeler]

The changes look reasonable to me. I thought there was a reason for it - the docs recommended the original method but I could be off on that. As long as the tests pass 👍

[toloudis]

The changes look reasonable to me. I thought there was a reason for it - the docs recommended the original method but I could be off on that. As long as the tests pass 👍

The bounds are definitely larger (1856) for a multichannel request because of that offset. ImageJ loads the data at 1848 (the actual plane size) instead of 1856. I haven't tested with bioformats. I feel comfortable that this code change is better for users who expect to just read the pixel data out of the czi file.