Fix RealisticColors compositor upcasting data to float64

[pnuu]

The RealisticColors compositor was upcasting the data to float64 (via np.where() usage). This PR fixes the issue, and adds the first test for the compositor.

• Tests added

[pnuu]

Creating the realistic_colors composite with main:

The same with this PR:

There's no change in memory usage, but the PR'd version is 17 seconds faster (20.0 s vs. 37 s).

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (fbe017c) 95.31% compared to head (60a7c1b) 95.35%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2669      +/-   ##
==========================================
+ Coverage   95.31%   95.35%   +0.03%     
==========================================
  Files         371      371              
  Lines       52435    52451      +16     
==========================================
+ Hits        49980    50013      +33     
+ Misses       2455     2438      -17     

Flag	Coverage Δ
behaviourtests	4.16% <0.00%> (-0.01%)	⬇️
unittests	95.96% <100.00%> (+0.03%)	⬆️

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 7138733453

• 20 of 20 (100.0%) changed or added relevant lines in 2 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.03%) to 95.916%

---

Totals
Change from base Build 7106990647:	0.03%
Covered Lines:	50139
Relevant Lines:	52274

---

💛 - Coveralls

[djhoese]

How did this never have tests before? Looks good to me. Thanks for fixing it.

One confusion: how does avoiding upcasting change the execution time so much?

[pnuu]

One confusion: how does avoiding upcasting change the execution time so much?

My guess is that the tasks now interleave better with each other. The same can be observer with most of the other dtype and premature computation fixes I've made so far.