[LPD][Correct] Fix/constant dtype conversion

Description

Before !867 (merged) and !779 (merged), constant arrays were loaded in the notebook and were initialized in the correct type for bad pixels.

After using CalibrationData API, https://git.xfel.eu/calibration/pycalibration/-/blob/master/src/cal_tools/calcat_interface.py#L626-638. Masks were stored in the wrong expected values for the corrected data.

with h5py.File(self.caldb_root / mdata["path"], "r") as cf:
    shape = cf[f"{dataset}/data"].shape
    dtype = cf[f"{dataset}/data"].dtype

As there are some old BadPixels constants that are stored as <f8 with NANs. When converting these nans to uint32, instead of being converted to 0 uint32 it gets converted to 2147483648

For the BadPixels, I convert NANs to zeros first. This bug was caught using the automated tests and the masks were the only non-validated data.

How Has This Been Tested?

Relevant Documents (optional)

Types of changes

• Bug fix (non-breaking change which fixes an issue)

Checklist:

Reviewers

@kluyvert

added Waiting for review label

assigned to @ahmedk

changed the description

Thanks, that LGTM. I wondered about doing this only if it's a float dtype, but it's probably not worth the extra code.

added 1 commit

• 170e231d - Avoid converting nans for uint32 badpixels

Compare with previous version

Thank you Thomas for the review!

changed target branch from feat/lpd_fragment2 to master

added 6 commits

• 170e231d...9dafd4fa - 4 commits from branch master
• 86fa6eb5 - convert f8 nan before converting to uint32
• 224f8ec9 - Avoid converting nans for uint32 badpixels

Compare with previous version

enabled an automatic merge when the pipeline for 224f8ec9 succeeds

resolved all threads

merged

mentioned in commit f3027732

changed milestone to %3.11.0