JP-3793: Fix numbers of flagged groups

[drlaw1558]

Resolves JP-3793 regarding inaccurate calculation of the number of groups in the jump step. This PR updates prior #317 to account for recent unrelated changes.

EDIT: See also JP-3877. This PR ensures that long TSO exposures don't fall into the wrong processing type, reducing runtime from over an hour to about 1 minute in the example given there.

[codecov]

Codecov Report

Attention: Patch coverage is 98.30508% with 1 line in your changes missing coverage. Please review.

Project coverage is 86.96%. Comparing base (9fba297) to head (f1278d7).
Report is 11 commits behind head on main.

Files with missing lines	Patch %	Lines
src/stcal/jump/twopoint_difference.py	95.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #338      +/-   ##
==========================================
+ Coverage   86.78%   86.96%   +0.18%     
==========================================
  Files          57       57              
  Lines       10401    10271     -130     
==========================================
- Hits         9026     8932      -94     
+ Misses       1375     1339      -36     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[melanieclarke]

Comparing to the previous PR, it looks like this is a clean copy of the original changes, also incorporating comments from Ken's review.

There are a couple errors in the unit tests that need cleaning up. When that's done, I can run regression tests.

[kmacdonald-stsci]

I recommend using parenthesis, instead of \ for multiple lines.

[drlaw1558]

Swapped line formatting.

[melanieclarke]

Regtests here:
jwst: https://github.com/spacetelescope/RegressionTests/actions/runs/13294000215
romancal: https://github.com/spacetelescope/RegressionTests/actions/runs/13294013308

[melanieclarke]

Romancal tests are okay, as expected - the failures are unrelated.

The jwst tests show some diffs for two of our regtest data sets, downstream of jump -- a NIRISS image and a NIRCam image. Since this is a bug fix, that's probably expected. In general, it looks like it found fewer jumps than before in both images. Are these diffs what you were expecting @drlaw1558?

[drlaw1558]

@melanieclarke It's not surprising, but I'll take a look at the test cases just to make sure.

[drlaw1558]

Ok, I’ve compared the rateint files for each of the two failing test cases. There are some number of pixels scattered throughout the images that vary visibly. It’s really low level though- some pixels improve, others get worse. On balance though, it looks like a significantly larger fraction of pixels improve with this PR than get worse (consistent with them now being processed through the appropriate branch of the jump step). I’d say that these differences are ok.

[melanieclarke]

Thanks for checking the diffs @drlaw1558.

This all looks good to me. Romancal folks should not be impacted by jump step changes, but I'm tagging them for review for their information.

Edit: I reran the romancal regtests this morning and the unrelated failures are now cleaned up - all tests pass.

[schlafly]

Regression tests pass & we don't use the stcal jump step; approving.