Introduce r-norms on manifolds with components

[kellertuer]

For manifolds that consist of components, like product, power and default manifold, this introduces an additional parameter to distance to specify the outer norm to be used after computing the single components distances; this especially also covers the 1- and ∞-norm cases.
It also introduces a has_components function to be able to check whether a manifold can compute such an (outer) norm.

🛣️🗺️

• ProductManifold
• PowerManifold
• DefaultManifold
• adapt the idea also to norm
• Documentation
• News and version number
• Testing / Code coverage

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.94%. Comparing base (7b52f31) to head (577a5ad).
Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #206      +/-   ##
==========================================
+ Coverage   99.90%   99.94%   +0.03%     
==========================================
  Files          31       31              
  Lines        3301     3398      +97     
==========================================
+ Hits         3298     3396      +98     
+ Misses          3        2       -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[mateuszbaran]

A few comments.

[kellertuer]

I rephrased the docs and made more clearer, that r-norm only applies to the outer layer.

[kellertuer]

Thanks for already fixing all the types (::Real). and defaults (to 2) – I hope I now found the last few tricky boundary cases, where functions are defined to avoid ambiguities.

Let's postpone the idea distance(M, p, q, m, Inf; recursive=true) to a later point. It could surely check whether any inner manifold has_components and for those cases pass the r down to them as well. But that would involve so many places I think (over 20?) that this would take me a day or two to find all the details. That is maybe not worth the effort for now, when I do not see who could find that useful.

[mateuszbaran]

Thanks! Let's not do recursive r in this PR, or maybe at all.

[kellertuer]

Can I merge this later today? I would like to maybe do the gradient change stopping criterion later today still, which was the origin for this feature.

[mateuszbaran]

Sure, I just have a few minor comments.