Remove direct estimagic dependencies to avoid circular imports

[janosg]

Remove direct estimagic dependencies to avoid circular imports

• Use scipy minimize in optimal sampling
• Write custom process batch evaluator
• Hardcode some algo options
• Look at benchmarks

The benchmarks change slightly. This is expected, as the optimization in the optimal sampler changed:

• scipy numerical derivative instead of estimagics's
• slightly different convergence criteria

I think we can ignore this. The changes are smaller than differences we get between computers.

Old parallel ls

New parallel ls

Old noisy ls

New noisy ls

Old ls

New ls

Old scalar

New scalar

[codecov]

Codecov Report

Merging #6 (43735cd) into main (4bce6be) will decrease coverage by 0.06%.
The diff coverage is 80.85%.

@@            Coverage Diff             @@
##             main       #6      +/-   ##
==========================================
- Coverage   77.96%   77.91%   -0.06%     
==========================================
  Files          69       68       -1     
  Lines        4852     4881      +29     
==========================================
+ Hits         3783     3803      +20     
- Misses       1069     1078       +9     

Impacted Files	Coverage Δ
src/tranquilo/tranquilo.py	97.56% <ø> (-0.06%)	⬇️
src/tranquilo/wrap_criterion.py	81.25% <58.33%> (-13.99%)	⬇️
src/tranquilo/process_arguments.py	94.30% <60.00%> (-3.06%)	⬇️
src/tranquilo/options.py	97.88% <100.00%> (ø)
src/tranquilo/sample_points.py	98.34% <100.00%> (+0.08%)	⬆️
tests/test_fit_models.py	100.00% <100.00%> (ø)
tests/test_tranquilo.py	100.00% <100.00%> (ø)
tests/test_visualize.py	100.00% <100.00%> (ø)

[timmens]

Looks good to me. I only have one question about an estimagic import, but I've already approved it.