GitLab

Aside from making the GitLab CI template more readable,
this PR also prevents me from comitting unresolved merge
conflicts unnoticed ever again.

this should minimize the pipeline display again since
gitlab should collapse more job names

[ci] fix weekly jobs on main

[algorithms.samdp] rename which argument

This PR  adds sphinxcontrib-bibtex to dependencies, converts bibliography to bibtex from rst and uses the ``:cite:`KEY` `` sphinx role instead of the `[KEY]_` rst feature.

This PR has two benefits:
  1. New references can be dropped into `bibliography.bib` without change
  2. The MyST parser cannot translate the `[KEY]_` syntax

This seems to be exactly what discussed during the developer meeting.

numpy 1.20 deprecated `np.float` and advises 
to either use float or np.float64

```
find . -name "*.rst" -o -name "*.py" | xargs sed -i  -E
's/(\[([^)]+?)\]\_)/:cite:`\2`/'
```
Which failed multiple matches on the same line, but those were only 4
files

This workaround treats the result of pytest as the success/fail trigger for the pipeline
not the exit code of the mpirun script.

The background here are intermittent failures were one rank does not call `MPI_Finalize` 
before exiting. While I made it configurable for pyMOR to tell mpi4py to do that
or not on module de-init, mpirun still sometimes errors out with "unclean"
process exits. These are never reproducable (locally or on pipeline re-run), not limited to 
python versions, nor have any other descernible pattern. Which is why I propose this workaround. 

This PR also forces the mpi runtime to use the in memory transport which
should stop the "slow transport fallback" message from popping up in logs.

Fix bug in apply_inverse for LincombOperators

This PR changes no functionality in pyMOR itself. 
But since the changes for CI images are substantial
I want to check and merge this update early, 
before adding any new code, so
that the images do not bit rot. 

While looking for some inspiration for the implementation of spacetime meshes, I stumbled across this small bug.

Limit variability and fix qt/ogl in azure CI