Fix CI and correct the setup.json

[sphuber]

Various fixes for the CI workflow

• Update version of actions checkout and setup-python to v2
• install Postgres and RabbitMQ through services. Note that we still
  need to install Postgres manually as well for pgtest to work.
• Remove manual install of numpy which was a temporary workaround
  that is now obsolete.
• Address the warnings from pylint
• Fix the pre-commit job: multiline run statement did not use |
• Add Python 3.9 to the tests matrix

Also the setup.json was

• Remove compatibility for Python 3.5 from classifiers as it is EOL.
  The tests were also not being run for this version, as the dependency
  aiida-testing does not support it.
• Add compatibility for Python 3.9 to classifiers
• Add the python_requires>=3.6 key
• Install aiida-core with the atomic_tools extra. This allows to
  remove the explicit installs of ase, pymatgen and PyCifRW.
• Add temporary upper limit on sqlalchemy which is a workaround for
  incompatibility with sqlalchemy-utils.
• Add upper limit on masci-tools because of breaking changes.
• Remove upper limit of numpy which is not necessary and causes
  conflicts with other packages for aiida-common-workflows.

[sphuber]

This PR supersedes #109 because the CI was broken and I needed to fix quite a few additional problems to get all tests to pass.

@broeder-j and @Tseplyaev I think these changes should fix the issue that we have in installing aiida-fleur with aiida-common-workflows in the Quantum Mobile. Do you think it would be ok to merge these changes and make a release? Or did I misunderstand your reasoning for putting an upper limit on numpy? I see that Vasily set the upper limit in this commit d3a64d0 but the message didn't specify why.

[codecov]

Codecov Report

Merging #110 (9eed552) into master (df33e9a) will increase coverage by 0.00%.
The diff coverage is 22.22%.

@@           Coverage Diff           @@
##           master     #110   +/-   ##
=======================================
  Coverage   68.56%   68.57%           
=======================================
  Files          71       71           
  Lines       10501    10501           
=======================================
+ Hits         7200     7201    +1     
+ Misses       3301     3300    -1     

Impacted Files	Coverage Δ
aiida_fleur/workflows/create_magnetic_film.py	25.00% <0.00%> (ø)
aiida_fleur/workflows/dmi.py	42.33% <0.00%> (ø)
aiida_fleur/workflows/eos.py	64.70% <0.00%> (ø)
aiida_fleur/workflows/mae.py	25.00% <0.00%> (ø)
aiida_fleur/workflows/mae_conv.py	25.80% <0.00%> (ø)
aiida_fleur/workflows/ssdisp.py	44.44% <0.00%> (ø)
aiida_fleur/workflows/ssdisp_conv.py	23.62% <0.00%> (ø)
aiida_fleur/tools/common_fleur_wf.py	83.48% <100.00%> (+0.31%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update df33e9a...9eed552. Read the comment docs.

[broeder-j]

Hi @sphuber. Thanks for doing this. I can only suspect that the upper limit on numpy at that time was due to dependency clashes with numpy, ase, pymatgen and bokeh, but I do not know. Since these are no major changes and so far there is no reason for the upper limit now anymore that I see and this is holding back the common-workflows, I am fine with a quick merge and release ASP after checking back with @Tseplyaev .