MAINT: Drop redundant travis configs

[bashtage]

Drop configs that are run on appveyor

• code/documentation is well formatted.
• properly formatted commit message. See
  NumPy's guide.

[codecov]

Codecov Report

Merging #5950 into master will decrease coverage by <.01%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #5950      +/-   ##
==========================================
- Coverage   82.96%   82.96%   -0.01%     
==========================================
  Files         600      600              
  Lines       95064    95009      -55     
  Branches    10460    10444      -16     
==========================================
- Hits        78871    78821      -50     
+ Misses      13777    13771       -6     
- Partials     2416     2417       +1

Impacted Files	Coverage Δ
statsmodels/tsa/stattools.py	86.79% <0%> (-1.25%)	⬇️
statsmodels/graphics/tests/test_mosaicplot.py	100% <0%> (ø)	⬆️
statsmodels/tools/grouputils.py	69.6% <0%> (+0.44%)	⬆️
statsmodels/regression/process_regression.py	77.61% <0%> (+1.16%)	⬆️

---

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 ae17fbd...ad6a650. Read the comment docs.

[coveralls]

Coverage decreased (-0.001%) to 85.502% when pulling 8c9587b on bashtage:update-travis into 95d094f on statsmodels:master.

[coveralls]

Coverage decreased (-0.01%) to 85.505% when pulling ad6a650 on bashtage:update-travis into ae17fbd on statsmodels:master.

[thequackdaddy]

So the travis pip version uses pyenv to get the python version and it fails because "3.6" is too ambiguous.

I noticed that the travis pip/openblas tended to be more delicate than the mkl version--especially for Mac. So I liked having separate ones. Not married to it, however.

[bashtage]

I think this can be dropped since python/pip is checked on azure for 3.5-3.7.

[bashtage]

I think this can be dropped since azure runs python/pip on OSX. The only issue here is that OpenBlas is tied to the NumPy version, so different NumPy will get different OpenBlas.