Refactor vectorfield kwargs

[janfb]

fixing kwargs issues in VF methods and adds a couple of edge case fixes for the time schedules.

See #1775 for context

closing #1775

Note: I introduce a new config dataclass for all VF parameters. it is used only internally and basically only for validating user-passed args and raising errors on typos or unknown args. Alternatively, we could do this manually by defining a dict or sequence of known args for each option, but I like the dataclass approach better and we could eventually use it a public config at a later point.

[codecov]

Codecov Report

❌ Patch coverage is 93.50649% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 87.77%. Comparing base (d41efa6) to head (aa3e824).
⚠️ Report is 1 commits behind head on main.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
sbi/inference/trainers/vfpe/npse.py	57.14%	3 Missing ⚠️
sbi/neural_nets/__init__.py	0.00%	1 Missing ⚠️
sbi/neural_nets/estimators/score_estimator.py	94.11%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1777      +/-   ##
==========================================
+ Coverage   87.65%   87.77%   +0.12%     
==========================================
  Files         140      140              
  Lines       12178    12224      +46     
==========================================
+ Hits        10675    10730      +55     
+ Misses       1503     1494       -9     

Flag	Coverage Δ
fast	83.42% <93.50%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
sbi/inference/trainers/vfpe/base_vf_inference.py	92.76% <100.00%> (-0.05%)	⬇️
sbi/inference/trainers/vfpe/fmpe.py	91.30% <100.00%> (+0.82%)	⬆️
sbi/neural_nets/estimators/base.py	74.60% <100.00%> (ø)
sbi/neural_nets/factory.py	93.24% <100.00%> (+11.31%)	⬆️
sbi/neural_nets/net_builders/vector_field_nets.py	94.03% <100.00%> (+0.72%)	⬆️
sbi/neural_nets/__init__.py	60.00% <0.00%> (+0.90%)	⬆️
sbi/neural_nets/estimators/score_estimator.py	91.89% <94.11%> (+0.35%)	⬆️
sbi/inference/trainers/vfpe/npse.py	84.61% <57.14%> (-5.87%)	⬇️

[manuelgloeckler]

Great, makes sense to change this!

I think one can do a few changes (see comments), especially I would adapt the downstream code to now actually use the introduced dataclass instead of the kwargs.

[manuelgloeckler]

Great!

Maybe it is already an issue, but I feel like a it would make sense to rewrite all the factories like this (i.e. replace the kwargs with dataclasses internally) so maybe makes sense to add an issue for that.