CircleCI -> Azure

[TomAugspurger]

Closes #22396
closes #22763

cc @jreback @h-vetinari @jorisvandenbossche @WillAyd

The main thing we lose is redundancy on the sql tests. Those are down to just travis.

[jschendel]

small nitpick: http --> https

[codecov]

Codecov Report

Merging #22992 into master will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master   #22992   +/-   ##
=======================================
  Coverage   92.22%   92.22%           
=======================================
  Files         169      169           
  Lines       51259    51259           
=======================================
  Hits        47275    47275           
  Misses       3984     3984

Flag	Coverage Δ
#multiple	90.66% <ø> (ø)	⬆️
#single	42.23% <ø> (ø)	⬆️

---

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 cd16b91...73c2413. Read the comment docs.

[TomAugspurger]

To summarize the changes:

Adds 3 new azure builds

1. 2.7-compat
2. 3.6-locale
3. 3.6-local-slow

That puts us up to 6 azure builds.

The 3.5 build is dropped. I think that's fine for now.

The locale stuff isn't working yet. Still debugging.

[TomAugspurger]

OK I think the locale setting took effect. If you look at the summary in https://dev.azure.com/pandas-dev/pandas/_build/results?buildId=656&view=ms.vss-test-web.test-result-details and search for a locale-related test, e.g. test_guess_datetime_format_with_locale_specific_formats, you'll see the skip message is Specific locale is set zh_CN.

I manually cancelled the travis build FYI, just to not overload it, it didn't fail on it's own.

edit: actually, restarting the travis build since I did have to change script-single and script-multi. For whatever reason the linux-based azure workers were having trouble publishing results from /tmp/*.xml. We got an insufficient privileges error, so maybe there's another xml file in there we can't touch. The solution was to just output the junitxml to our directory.

[jreback]

would be ok leaving the 1 circle ci test

[jbrockmendel]

GH wants a \n here

[jbrockmendel]

Needs rebase.

If circleCI isn't completely removed, it might be nice to collect the relevant scripts in ci/circleci/

[TomAugspurger]

Restored one circle job for the SQL redundancy.

I'd recommend that if circle is running behind, we feel free to ignore it for changes that aren't likely to affect the SQL behavior.

[TomAugspurger]

This is passing.

I'm not thrilled about having to maintain the circleci scripts, but it's probably worth it for the SQL redundancy.

[jreback]

@TomAugspurger I think i merged something in contributing.rst so rebase and let's merge on green.

[TomAugspurger]

Apparently, circleCI missed the web hook on my latest one. But it did pass in https://circleci.com/gh/pandas-dev/pandas/26751?utm_campaign=vcs-integration-link&utm_medium=referral&utm_source=github-build-link, which only differed by the doc fixup, so merging.