Refactor create_cache

[JoshuaLampert]

This passes the boundary_conditions to create_cache and dispatches on them instead of looking at the type of the SBP operators to infer whether the problem has periodic or reflecting boundary conditions, e.g. previously it was possible to use boundary_conditions_reflecting and pass periodic operators to the Solver, which meant that the create_cache function also created the cache for the periodic problem instead of for reflecting. This was a bit messy. This way it should work more as expected and is also easier to extend.
On the fly, I put tmp1(needed by the RelaxationCallback) into the intitial_cache instead of always creating it in create_cache, which is weird because tmp1 is sometimes not needed for the discretization itself.
Finally, I fixed the upwind discretization for the reflecting boundary conditions for the BBMBBMEquations1D, which was broken before. I also added a test in a similar manner as in #105.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 96.71%. Comparing base (6d27850) to head (87611ae).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #106      +/-   ##
==========================================
+ Coverage   96.44%   96.71%   +0.26%     
==========================================
  Files          17       17              
  Lines        1125     1126       +1     
==========================================
+ Hits         1085     1089       +4     
+ Misses         40       37       -3     

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

[ranocha]

Thanks! I just have some minor comments that can either be resolved quickly or maybe postponed to later development (issues). Feel free to merge 👍

[ranocha]

In the long run, one should consider allowing more flexibility here. For example, FD methods should use sparse matrices and just a factorization instead of the inverse of a dense matrix. But that's not critical right now.

[JoshuaLampert]

Yeah, you are definitely right. IIRC, when I tried that some time ago, I had some problems making that properly work, but I don't remember the exact issue. Back then I focused on something that just works and is easy to implement and understand. But of course this is not the most performant way. I created #107 for that.