MAINT: Manage versioning with versioneer

[effigies]

This is a follow-up to #764. Now that we have a tag post-upstream/2.5.0.dev, Versioneer should be able to construct valid PEP-440-style version strings from any revision, and we can drop our custom githash handling code and nisext.

I don't think this should have negative impacts, but I'm going to leave this one open for a week or two in case anybody foresees problems with this move.

[effigies]

This ended up leading to a sort of chicken-egg problem with long_description. Because versions are no longer set statically, nibabel/info.py is no longer exec-able, and we can't move __version__ into __init__.py because of the cmp_pkg_version function. So I ended up just moving the canonical text of long_description into a long_description.rst file in the root, and updating refresh_readme.py accordingly.

[codecov]

Codecov Report

Merging #786 into master will decrease coverage by <.01%.
The diff coverage is 83.78%.

@@            Coverage Diff            @@
##           master    #786      +/-   ##
=========================================
- Coverage   90.11%   90.1%   -0.01%     
=========================================
  Files          94      95       +1     
  Lines       11886   11879       -7     
  Branches     2119    2118       -1     
=========================================
- Hits        10711   10704       -7     
- Misses        831     834       +3     
+ Partials      344     341       -3

Impacted Files	Coverage Δ
nibabel/deprecated.py	100% <100%> (ø)	⬆️
nibabel/_version.py	100% <100%> (ø)
nibabel/info.py	100% <100%> (ø)	⬆️
nibabel/__init__.py	92.85% <100%> (+0.12%)	⬆️
nibabel/pkg_info.py	81.81% <79.31%> (+9.4%)	⬆️
nibabel/openers.py	90.81% <0%> (-4.09%)	⬇️
nibabel/viewers.py	96.05% <0%> (+0.65%)	⬆️

---

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 7c69442...30c536f. Read the comment docs.

[effigies]

One remaining question is whether to move from the custom version comparison code to packaging.version.Version, which is the reference implementation of PEP440. Our comparison is not PEP440-compatible, as it assumes that any post-string means a pre-release version.

It would be a new dependency, but fairly lightweight and pushes correctness concerns upstream.

[effigies]

No major outcry, and we're testing git archive now, so I think all of the old behavior is tested and has a decent fallback option. I don't think I'll bother trying to port this to 2.5.x, unless anybody feels strongly about consistent versioning for all active branches.