FLSHCORR: Scale for linear decrease over time

[pllim]

FLSHCORR: Scale for linear decrease over time if the reference image is the last ever to be delivered.

But we do not want to merge this until we know for sure, though I coded it up such that it would still run without the scaling if those new keywords are not found in the reference image primary header.

Fix #627

RT https://github.com/spacetelescope/RegressionTests/actions/runs/14917237097

TODO

• Run CI and RT, and fix any failures.
• Wait for ACS Team sign off to merge or not. Rerun CI and RT.
• Add new test case with prototype reference image to regression test.

[pllim]

As of 2025-05-08 implementation, if you give it the existing FLSHFILE from CRDS without these new keywords, it would skip the scaling (the following output is from -v option, so you call calacs.e blah_raw.fits -v). It would report the keyword is missing but then keep going without the scaling. It would also stop checking at the first missing keyword (3 total):

FLSHCORR PERFORM
Trying to open jref$39h2109kj_fls.fits...
Read in Primary header from jref$39h2109kj_fls.fits...
ERROR:    Keyword = `SLOPE'.
Performing post-flash subtraction on chip 1 in imset 2
Linear decrease over time scaling for post-flash: SKIPPED
Image has an offset of 0,0
Mean of post-flash image (MEANFLSH) = 17.0107
FLSHCORR COMPLETE

And if you give it the new FLSHFILE, it would be like this instead:

FLSHCORR PERFORM
Trying to open test_fls.fits...
Read in Primary header from test_fls.fits...
Performing post-flash subtraction on chip 1 in imset 2
Linear decrease over time scaling for post-flash: 1.026984e+00
Image has an offset of 0,0
Mean of post-flash image (MEANFLSH) = 12.2674
FLSHCORR COMPLETE

[pllim]

p.s. The scaling factor is greater than 1 probably because I just grabbed a random image from Artifactory to test and it is very old.

[mdlpstsci]

Code changes look fine.

[pllim]

Update 2025-05-21: Jenna gave me a couple of sample inputs and expected outputs. I ran ACS2D from this branch and compared my outputs to hers. There are differences and I am waiting to hear back if they are expected or not.

[pllim]

At ACS Team meeting, we decided that scaling for old data is not pipeline problem because CRDS should not use this last flash file for old data where better match is available. If user purposefully stick this last flash in their old data, they are asking for it.

Local testing of this branch with sample inputs adjusted for years 2025 and 2030, a sample last flash reference file, and expected output values gracefully provided by Jenna Ryon showed that this branch implementation is good.

I will keep this in draft and revisit when we have a "go" from ACS Team.