Stephan Rave

@artpelling, can this be closed in favor of #1952?

> 2\. Main has [MPL](https://docs.pymor.org/main/tutorial_builtin_discretizer.html) output still, so I wouldn't say it's a hard blocker. Ah, sorry, didn't see that this PR also removes the pin on ipywidgets. Then I'd...

Seems like I should complain more often ... Maybe we can discuss things offline next week?

> The pipelines are failing due to a very old scipy version (1.5.4) where the 'highs' method (which is now the default) was not available as solver of a linear...

@HenKlei, could you also update the first PR comment and make it a bit more informative?

This is a bit more difficult than I initially thought: While returning the Newton stages via the `compute` output dict is no problem, `StationaryModel._compute_solution` calls `operator.apply_inverse`, which then dispatches into...

@artpelling, any ideas why the your tests seem to be hanging in CI. Have you tried locally?

@artpelling, @pmli, tests are really hanging because `shift` is computed to be 0 as the norm of XX is. This could be mitigated by setting the norm to 1 in...

> > However, if the input data is zero, this will still lead to an 'orthonormal' basis of zero vectors. I'd say the only practical approach would be to fail...

Regarding testing, I'd suggest to call `gram_schmidt` in the test with some reasonable defaults to detect if the input array is rank deficient. If it is, one can assert that...