[PLS Refactor] Linear solver stopping criteria

[JonathanWenger]

In a Nutshell

This PR adds stopping criteria for the probabilistic linear solver.

Detailed Description

The following stopping criteria are added in addition to an abstract base class:

• MaxIterationsStopCrit: converge after a maximum number of iterations
• ResidualNormStopCrit: converge for a sufficiently small residual norm
• PosteriorContractionStopCrit: converge for sufficiently small trace of the posterior covariance

Related Issues and PRs

PR is part of #51 and supersedes part of #283.

[codecov]

Codecov Report

Merging #499 (a836666) into main (879cdfe) will increase coverage by 0.10%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #499      +/-   ##
==========================================
+ Coverage   87.62%   87.72%   +0.10%     
==========================================
  Files         155      160       +5     
  Lines        5963     6012      +49     
  Branches      771      772       +1     
==========================================
+ Hits         5225     5274      +49     
+ Misses        505      504       -1     
- Partials      233      234       +1     

Impacted Files	Coverage Δ
.../solvers/information_ops/_linear_solver_info_op.py	100.00% <ø> (ø)
.../probnum/linalg/solvers/information_ops/_matvec.py	100.00% <ø> (ø)
...m/linalg/solvers/information_ops/_proj_residual.py	100.00% <ø> (ø)
...obnum/linalg/solvers/stopping_criteria/__init__.py	100.00% <100.00%> (ø)
...ping_criteria/_linear_solver_stopping_criterion.py	100.00% <100.00%> (ø)
...obnum/linalg/solvers/stopping_criteria/_maxiter.py	100.00% <100.00%> (ø)
...olvers/stopping_criteria/_posterior_contraction.py	100.00% <100.00%> (ø)
...linalg/solvers/stopping_criteria/_residual_norm.py	100.00% <100.00%> (ø)
...um/linalg/solvers/beliefs/_linear_system_belief.py	76.92% <0.00%> (ø)
... and 4 more

[JonathanWenger]

Note that this PR is a follow-up to #494 and should therefore be reviewed after.

[marvinpfoertner]

@JonathanWenger I merged main into this branch to facilitate the review. I hope that was fine with you.

[marvinpfoertner]

👍🏻

[marvinpfoertner]

We might want to cache this

[JonathanWenger]

As we previously discussed offline, I would avoid caching the residual or right-hand side at this point, before we actively need it for performance. We may want to either recompute the residual in each step for stability or update the (estimate of) right hand side at each step in the noisy case. Let me know if your opinion on this is different.