Run CI on GitHub Actions

[elacuesta]

The idea is to stick to the title and avoid overlapping with #407 as much as possible, but a few things might be duplicated (Flake8 adjustments and Python classifiers, for instance).

• Flake8 job (small adjustments)
• Tests (Python 3.6 to 3.9; Ubuntu, macOS, Windows)
• Readme badges
• Removed Python 2.7 and 3.5 classifiers
• PyPI publishing job
• Add PyPI token
• Attach artifacts to (draft) release (binaries created with the "freeze" tox environment)

Commit history is not very clean, it should be squash-merged.

[codecov]

Codecov Report

Merging #413 (908c0a1) into master (51a53b0) will decrease coverage by 1.94%.
The diff coverage is n/a.

❗ Current head 908c0a1 differs from pull request most recent head 1b668e4. Consider uploading reports for the commit 1b668e4 to get more accurate results

@@            Coverage Diff             @@
##           master     #413      +/-   ##
==========================================
- Coverage   91.42%   89.48%   -1.95%     
==========================================
  Files          34       34              
  Lines        2415     2415              
==========================================
- Hits         2208     2161      -47     
- Misses        207      254      +47     

Impacted Files	Coverage Δ
shub/deploy_egg.py	82.53% <0.00%> (-14.29%)	⬇️
shub/compat.py	52.17% <0.00%> (-13.05%)	⬇️
shub/utils.py	82.39% <0.00%> (-6.77%)	⬇️

---

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 5fcb15f...1b668e4. Read the comment docs.

[elacuesta]

This test works when executed on its own, but fails when executed as part of the whole test suite. While it would be great to know exactly what causes this, at this point I think it's more important to restore CI to a functioning state.

[Gallaecio]

The code looks good to me, my only concern is one I expressed in #407:

I’m not sure about dropping Python support. I think it may be safer to do this after the rebranding.

It may not be a valid concern anymore, though.

Does Scrapy Cloud still support those removed Python versions? If so, we could still go ahead with this change, but we should probably include the python version requirement into setup.py, so that users with a modern pip get a version of shub compatible with Python 2, and also update documentation resources about installing shub to take into account the possibility that users have an old pip and hence they should manually pin the shub version in the pip install command.