Refactor tranquilo code

[janosg]

• Improve construction of test cases in test_tranquilo.py.
• Make sure we have one function to calculate the necessary number of parameters/degrees of freedom for linear and quadratic models. Should be defined in models.py.
• Remove unused arguments in the aggregator interface.
• Tests for input processing functions in tranquilo.py.

[codecov]

Codecov Report

Merging #395 (e00d302) into main (4b9019f) will increase coverage by 0.10%.
The diff coverage is 96.34%.

@@            Coverage Diff             @@
##             main     #395      +/-   ##
==========================================
+ Coverage   92.62%   92.73%   +0.10%     
==========================================
  Files         208      209       +1     
  Lines       16372    16484     +112     
==========================================
+ Hits        15165    15286     +121     
+ Misses       1207     1198       -9     

Impacted Files	Coverage Δ
src/estimagic/optimization/tranquilo/fit_models.py	83.73% <57.14%> (+0.19%)	⬆️
.../estimagic/optimization/tranquilo/filter_points.py	80.24% <66.66%> (+0.24%)	⬆️
src/estimagic/optimization/tranquilo/models.py	96.15% <93.75%> (-3.85%)	⬇️
...timagic/optimization/tranquilo/aggregate_models.py	87.71% <100.00%> (ø)
src/estimagic/optimization/tranquilo/tranquilo.py	95.85% <100.00%> (+5.38%)	⬆️
...ts/optimization/tranquilo/test_aggregate_models.py	100.00% <100.00%> (ø)
tests/optimization/tranquilo/test_filter_points.py	100.00% <100.00%> (ø)
tests/optimization/tranquilo/test_models.py	100.00% <100.00%> (ø)
tests/optimization/tranquilo/test_tranquilo.py	100.00% <100.00%> (ø)
...ptimization/subsolvers/bounded_newton_quadratic.py	86.40% <0.00%> (ø)
... and 2 more

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[timmens]

1. subsolvers/quadratic_subsolvers.py, subsolvers/bounded_newton_quadratic.py and models.py contain functions to evaluate models. However, they use different arguments. In subsolvers/linear_subsolvers.py a linear model is evaluated, but we cannot use the other functions straight aways since they assume that the input has at least the attribute square_terms.
2. Function _polynomial_features in fit_models.py is very similar to _square_features in filter_points.py. We could think about if we can combine these two or call one from the other. I did not find a way to efficiently / cleanly combine these two functions. It seems that for now we should keep both of them.

[timmens]

We will tackle problems 1. and 2. in separate PRs.

[segsell]

Very nice refactoring PR! The test modules look way cleaner now. And great job finding out that np.linalg.lstsq is more robust than np.linalg.sovle in powell_fitting. Thank you!!

[segsell]

The powell fitting benchmarks (scalar case) now look slightly worse switching from np.linalg.solve to np.linalg.lstsq. But this is hardly noticeable in the relevant case of 2 * n + 1 points.