Fix dtype promotion in SunZenithReduction

[pnuu]

The SunZenithReduction modifier used for example in the built-in FCI true_color composite has a division with np.log(2) that changes the float32 input data to float64. This PR is a simple fix for that.

The upcasting is hidden behind some other part of the code because even without this fix the completed product was float32.

• Tests added

This small change also reduced the memory usage for my 12-composite set by roughly 800 MB.

With main:

With this PR:

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 96.08%. Comparing base (f36a096) to head (6193040).
Report is 28 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2950   +/-   ##
=======================================
  Coverage   96.08%   96.08%           
=======================================
  Files         377      377           
  Lines       55125    55144   +19     
=======================================
+ Hits        52967    52986   +19     
  Misses       2158     2158           

Flag	Coverage Δ
behaviourtests	3.94% <0.00%> (-0.01%)	⬇️
unittests	96.18% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

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

[coveralls]

Pull Request Test Coverage Report for Build 11669217385

Warning: This coverage report may be inaccurate.

This pull request's base commit is no longer the HEAD commit of its target branch. This means it includes changes from outside the original pull request, including, potentially, unrelated coverage changes.

• For more information on this, see Tracking coverage changes with pull request builds.
• To avoid this issue with future PRs, see these Recommended CI Configurations.
• For a quick fix, rebase this PR at GitHub. Your next report should be accurate.

Details

• 17 of 17 (100.0%) changed or added relevant lines in 2 files are covered.
• 1 unchanged line in 1 file lost coverage.
• Overall coverage increased (+0.001%) to 96.192%

---

Files with Coverage Reduction	New Missed Lines	%
satpy/tests/writer_tests/test_cf.py	1	99.14%

Totals
Change from base Build 11634318306:	0.001%
Covered Lines:	53230
Relevant Lines:	55337

---

💛 - Coveralls

[pnuu]

The other two things doing upcasting for True color composite that hid this one were fixed in the merged #2954 and #2956 PRs.

[djhoese]

So this fixes some extra memory being used inside a dask task, but doesn't change the final dtype of the output of this "angles" step (which was correct), right?

Side note: Based on this math website, can't this equation be rewritten as log-base2 of reduction_factor + 1?

https://www.cuemath.com/algebra/log-rules/

change of base rule: loga b = (logc b) / (logc a)

[pnuu]

With the current main float32 input data resulted in output of float64 because of np.log(2) promoting reduction_factor to float64 and finally causing the same to res. With this fix the output will also be float32. float64 input will still yield float64 result.

[pnuu]

And well spotted, the log2 is much simpler in every way. Fixed in 6193040

[djhoese]

And well spotted, the log2 is much simpler in every way. Fixed in 6193040

Hopefully that is still mathematically equivalent and I'm not misreading it.

[pnuu]

And well spotted, the log2 is much simpler in every way. Fixed in 6193040

Hopefully that is still mathematically equivalent and I'm not misreading it.

It is, the tests pass (at least locally).

[pnuu]

I restarted CI, MacOS tests failed on setup.