Final Manylinux2014-based release (2.0.0-rc.2)

[ptheywood]

Due to the upcoming removal of support for github actions actions/upload-artifact@v3 & actions/download-artifact@v3, our wheel generation on release process will break after 2024-11-30.

v4 is incompatible with older GlibC releases, so we must drop Manylinux2014 support because of this. If we want a "final" release with support for Centos7, we need to do it in November 2024.

Prior to this, we should try and get a couple of things fixed / merged:

• SWIG 4.1.x, 4.2.x & 4.3x #956

  • Manylinux 2014 includes swig 4.2.1.post0 (at the time of writing)
  • This now produces usable wheels since Graph visualisation would only work if CUDASimulation was explicitly init() #1243 & Tests: Workaround for Swig 4.2.1 != None erorrs #1244, but generated wheels will still include the != None / == None issue in user-code.
  • Do atleast some of SWIG 4.1.x, 4.2.x & 4.3x #956 (comment) ?
    • Minimimum is to add it as a known issue to the release notes.

• cpplint 2.x #1241 via Cpplint 2 fixes #1245

  • This shouldn't block release, but would be nice if CI were happy.

• Remove unnecessary astpretty dependency #1164?

  • Not required, but would be nice to reduce python packaing deps if time allows.

• Release steps:

  • Vis repo tag + update hash
  • Changelog
    • include deprecated glibc < 2.28 / manylinux2014 due to CI changes / EoL for ManyLinux2014 base OS.
  • Citation.cff
  • Check produced wheels via CI work
  • Release notes
    • Include that this is the final release with manylinux2014 (semi) compatibly wheels, future releases will be ManyLinux_2_28 / require glibc >= 2.28. I.e. this removes centos7 support from binary wheels. Should still be fine from source.
    • Mention the swig == None issue if still releasing with swig 4.2.1.

After this release, but ideally before the end of november, we can then finish and merge #1228 (otherwise most workflows will start to fail).

The subsequent release could then include many of the outstanding fixes we don't have time for (and we should try and get nvrtc runtimes imrpoved for CUDA > 12.2)