Change actions for mamba / pip install methods.

[nbiederbeck]

for mamba: use mamba-org/provision-with-micromamba
for pip: use actions/setup-python

[StFroese]

Should we also update the installation instruction in the README and use mamba instead of conda?
I mean it states using mamba if solving environment takes to long but is there anything against telling the user to use mamba directly?

[kosack]

Eventually, mamba won't be needed explicitly as the mamba solver will be included in a future version of conda (I think it's experimentally there already), but for now I think it's fine to just recommend mamba, as we know that installing packages from conda-forge with conda can be a slow nightmare due.

[maxnoe]

I think we should always prefer recommending the open-source builds by conda-forge / mamba over the closed source (binary) blob packages by Anaconda Inc.

[nbiederbeck]

I also update codecov here, instead of #2206. I think it is reasonable to have this in a single PR.

[LukasNickel]

There has been a new matplotlib release yesterday (3.6.3), that the pip install uses, but mamba does not.
It is not immediately obvious to me however what change in matplotlib would cause the tests to fail now.
https://matplotlib.org/stable/users/github_stats.html

[kosack]

seems they have change the behavior of how vmin/vmax are calculated such that it breaks if there are NaN or Inf values? The test is failing when setting a CameraDisplay to log-scale, which will cause NaNs for any negative value (but that was not an issue in the past). Could also be a side effect from something in numpy