Fix test_argcomplete: use python -m pytest

[blueyed]

Previously it was not run with a) xdist ("-c"), and b) "python -m
pytest" ("…/pytest.py", not executable).

We still need an env to acutally have argcomplete installed.

Where should it go?
Thought about py27-trial (what does trial stand for there?), but it appears to only run test_unittests?!

[codecov]

Codecov Report

Merging #4848 into master will increase coverage by 0.09%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #4848      +/-   ##
==========================================
+ Coverage   94.25%   94.35%   +0.09%     
==========================================
  Files         113      113              
  Lines       25163    25160       -3     
  Branches     2498     2497       -1     
==========================================
+ Hits        23718    23739      +21     
+ Misses       1107     1088      -19     
+ Partials      338      333       -5

Flag	Coverage Δ
#linux	91.75% <50%> (+0.08%)	⬆️
#windows	93.46% <100%> (-0.01%)	⬇️

Impacted Files	Coverage Δ
testing/test_parseopt.py	97.97% <100%> (+3.17%)	⬆️
src/_pytest/terminal.py	91.74% <0%> (+2.64%)	⬆️

---

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 e711a6c...ccab469. Read the comment docs.

[blueyed]

Hmm, Travis does not run for draft PRs, but Azure does..

[nicoddemus]

Where should it go?

Hmmm can't we install argcomplete in every environment, like we do with hypothesis and mock? Seems like it shouldn't be a big dependency.

Thought about py27-trial (what does trial stand for there?), but it appears to only run test_unittests?!

"trial" because it tests the "twisted trial" functionality, which is the builtin facilities to test "twisted" applications. I agree it is strange, I believe it would be better to rename to py27-twisted instead.

[nicoddemus]

Now that I think about it, why we don't just install numpy and twisted in every environment as well? I think the cost of doing that would greatly offset the extra jobs we have just for those dependencies, specially with wheels available for numpy (which I suspect was the cause for splitting those environments in the first place).

[blueyed]

Sounds good, will do.