Fix ChambollePock TangentBundle methods

[mateuszbaran]

I'm fixing an issue introduced by JuliaManifolds/Manifolds.jl#509 , that is TangentBundle no longer implementing exp, log, distance and parallel_transpot. Most issues are fixed by just having separate options for tangent bundle in options but one remaining problem is that DebugEntryChange requires distance to be implemented, which in turn requires log. My idea to solve that is to introduce inverse retraction-dependent distance approximations in ManifoldsBase.jl, that is something like distance(M::AbstractManifold, p, q, m::AbstractInverseRetractionMethod) = norm(M, p, inverse_retract(M, p, q, m)). What do you think about it?

[codecov]

Codecov Report

Merging #138 (47c8bc3) into master (066b998) will not change coverage.
The diff coverage is 100.00%.

❗ Current head 47c8bc3 differs from pull request most recent head 32d33a9. Consider uploading reports for the commit 32d33a9 to get more accurate results

@@           Coverage Diff           @@
##           master     #138   +/-   ##
=======================================
  Coverage   98.70%   98.70%           
=======================================
  Files          51       51           
  Lines        3322     3322           
=======================================
  Hits         3279     3279           
  Misses         43       43           

Impacted Files	Coverage Δ
src/solvers/ChambollePock.jl	100.00% <ø> (ø)
src/plans/primal_dual_plan.jl	100.00% <100.00%> (ø)

📣 Codecov can now indicate which changes are the most critical in Pull Requests. Learn more

[kellertuer]

I think overall this looks fine and thanks for the fix, I am just a little insecure about the naming

inverse_retraction_method_tb but have not yet had a good idea how to name it instead. I feel _tb is not descriptive enough but writing tangentbundle is too long as an overall variable name.

[mateuszbaran]

I was also thinking about calling it inverse_retraction_method_m and inverse_retraction_method_n, maybe that would be fine?

[kellertuer]

That would be an idea, but a breaking change.

[kellertuer]

What about retraction_method_dual and ìnvere_retraction_method_dual` since it refers to the n (N) one for the dual variable?

[mateuszbaran]

That sounds fine 👍

[kellertuer]

We could shortly check that the examples still run fine (but the tests still do, so probably they'll do as well), but otherwise this looks nice.

[mateuszbaran]

I really doubt this change can break any examples.