Improve trust region sampling in tranquilo

[timmens]

In this PR we want to improve the trust region sampling of Tranquilo. In particular, we implement

Sphere sampling

• Naive (random points on the sphere)
• Optimal (points on the sphere placed to maximize the minimum distance between two points)

Cube sampling

• Naive (random points on the cube)
• Optimal (points on the sphere placed to maximize the minimum distance between two points)

If binding bounds are present we map the shape (sphere or cube) into the feasible trustregion, which is defined as the intersection of the bounds and the trustregion.

[codecov]

Codecov Report

Merging #388 (1b5c6a6) into main (e0b5b6c) will increase coverage by 0.05%.
The diff coverage is 96.98%.

@@            Coverage Diff             @@
##             main     #388      +/-   ##
==========================================
+ Coverage   92.73%   92.78%   +0.05%     
==========================================
  Files         209      209              
  Lines       16484    16628     +144     
==========================================
+ Hits        15286    15428     +142     
- Misses       1198     1200       +2     

Impacted Files	Coverage Δ
tests/optimization/tranquilo/test_filter_points.py	100.00% <ø> (ø)
.../estimagic/optimization/tranquilo/sample_points.py	91.05% <93.58%> (+1.77%)	⬆️
src/estimagic/optimization/tranquilo/options.py	100.00% <100.00%> (ø)
src/estimagic/optimization/tranquilo/tranquilo.py	96.04% <100.00%> (+0.18%)	⬆️
...imagic/optimization/tranquilo/tranquilo_history.py	75.53% <100.00%> (+0.53%)	⬆️
tests/optimization/tranquilo/test_sample_points.py	100.00% <100.00%> (ø)
...magic/optimization/subsolvers/_trsbox_quadratic.py	92.24% <0.00%> (+1.22%)	⬆️

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

[janosg]

Very nice!