Update neural network tests

[gdalle]

DIT extensions

• Use the Constant context mechanism for Flux and Lux scenarios.
• Make Flux scenarios reproducible with rng.
• Use ForwardDiff instead of FiniteDiff for the reference of Lux scenarios (doesn't work with Flux because of Flux.reset!). Keep ComponentArray on both sides for result comparison.
• Deactivate Flux scenarios with iterated model application, for some reason they're failing now.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.30%. Comparing base (097930a) to head (ab8f766).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #490      +/-   ##
==========================================
+ Coverage   97.87%   98.30%   +0.43%     
==========================================
  Files         107      107              
  Lines        4322     4311      -11     
==========================================
+ Hits         4230     4238       +8     
+ Misses         92       73      -19     

Flag	Coverage Δ
DI	98.31% <ø> (ø)
DIT	98.28% <100.00%> (+1.33%)	⬆️

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.