Make quad policies stateless

[mmahsereci]

In a Nutshell

The BQ policies were not stateless. This PR makes them stateless by passing the rng to the call function instead it being an attribute of the policy instances. The PR also adds some long overdue shape tests for the policies.

Detailed Description

should be reviewed after #743 is merged since it branches off of it.

Related Issues

Closes #...

[codecov]

Codecov Report

Merging #744 (7c0839d) into main (92e59ad) will increase coverage by 0.03%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #744      +/-   ##
==========================================
+ Coverage   90.51%   90.54%   +0.03%     
==========================================
  Files         202      202              
  Lines        7538     7552      +14     
  Branches      974      974              
==========================================
+ Hits         6823     6838      +15     
+ Misses        481      480       -1     
  Partials      234      234              

Impacted Files	Coverage Δ
src/probnum/quad/_bayesquad.py	100.00% <100.00%> (ø)
src/probnum/quad/solvers/_bayesian_quadrature.py	96.73% <100.00%> (+1.04%)	⬆️
src/probnum/quad/solvers/policies/_policy.py	100.00% <100.00%> (ø)
...rc/probnum/quad/solvers/policies/_random_policy.py	100.00% <100.00%> (ø)
...um/quad/solvers/policies/_van_der_corput_policy.py	100.00% <100.00%> (ø)
...bnum/quad/solvers/stopping_criteria/_max_nevals.py	100.00% <100.00%> (ø)

[JonathanWenger]

All looks good, except for the implicit use of the global random state. I think we should always force the user to provide their own random number generator to avoid unexpected side effects or non-determinism. With the changes coming in #581 this will become mandatory anyways.

[mmahsereci]

All looks good, except for the implicit use of the global random state. I think we should always force the user to provide their own random number generator to avoid unexpected side effects or non-determinism. With the changes coming in #581 this will become mandatory anyways.

Sure, I'll remove the default. What about wrappers like bayesquad that are not really part of the core library? I assume the default rng there needs to go as well @JonathanWenger ? And thanks for reviewing so promptly!

[JonathanWenger]

Yes, there the user should also provide their own RNG. It should still be optional and raise an error if at some point the compiled BQ method does need random numbers. Same as for the 'inegrate' method.