BUG: constrain cftime to >=1.1.1

[CameronTStark]

• Closes BUG: netcdf4 conversion error in forecast.py  #944
• I am familiar with the contributing guidelines
• Adds description and name entries in the appropriate "what's new" file in docs/sphinx/source/whatsnew for all changes. Includes link to the GitHub Issue with :issue:`num` or this Pull Request with :pull:`num`. Includes contributor name and/or GitHub username (link with :ghuser:`user`).
• Pull request is nearly complete and ready for detailed review.
• Maintainer: Appropriate GitHub Labels and Milestone are assigned to the Pull Request and linked Issue.

Limiting cftime<=1.1.0 . This is a few releases ahead of the suggested 1.0.4 but still passes all tests.

[CameronTStark]

It's good that everything checked out on this run. I reran the failing tests from #937 and they still fail confirming that this is indeed a fix.

I'll add to what's new and this should be ready for approval thereafter.

[CameronTStark]

This is ready for review. Edit: needs more work

This is by no means the solution and an issue to address needs to be drafted unless #920 encompasses it. I'll make the issue unless If someone who knows the inner workings of forecast.py and/or how this interacts with it can describe what needs to be done.

[CameronTStark]

As proof of the error I pushed after reverting this branch to master to get the test results. Link to test results

The error is fixed in Python3.6 and above (thanks @kanderso-nrel!) but persists in Python3.5 even with the changes. Link to 3.5 error

cftime only supports >=3.6 and therefore doesn't have the kwarg only_use_python_datetimes for the downloaded version in the 3.5 test.

The Test_conda_linux environment for Python3.5 installs 1.0.0b1. Link to installed packages for 3.5 test

All other Python versions install the most current 1.1.1 Link to installed packages for 3.6 test

The most obvious options are:

• Go back to pinning 1.1.0
• Drop support for Python3.5

Are there any more options?

[wholmgren]

Pinning is ok with me, but Another option is to xfail or skip the test if run on 3.5. I’d rather wait until 0.8 to drop 3.5.

[CameronTStark]

I think this, after much learning, is ready for review.

Thanks for your patience for the time it took and notification spam.

Please let me know if the codecov/patch failure needs to be addressed for test. I don't think so but there may be some historical context in which I'm unaware.

Edit:
I was also wondering if there should be a warning in get_data() for 3.5 users when this fails. If so, what's standard practice?

[wholmgren]

ok to ignore this codecov/patch failure. Ok with me to just let it break on python 3.5 if we're no longer supporting it in pvlib 0.8.

[CameronTStark]

When you get a moment @wholmgren can you please merge this? Thanks!