Fix pint 0.9 errors from units.wraps and iterable

[jthielen]

As discussed in #997, there were two main issues with pint 0.9 and MetPy: @units.wraps and single value Quantities being evaluated as iterable. This PR implements an alternate way of ensuring matching units (and stripping the units) for the interpolation functions that previously used @units.wraps, as well as an iterable function that checks against the magnitude of the argument.

The two tests with matplotlib and scalar Quantities remain broken. I marked them as skipped for pint 0.9, but if a monkey patch of some kind based on the fix discussed in hgrecco/pint#751 is preferred, I can see what can be done.

Fixes #997.

[dopplershift]

Looks good, pending CI.

[jthielen]

Pint 0.9 is still failing with Python 2.7 on Travis...if I'm seeing things correctly, @dopplershift your fix from January was merged into pint master, but still hasn't made it to release yet on pip (even though it is fixed on conda-forge...there's an open issue on pint for this: hgrecco/pint#819).

So, what should be done?

(Yet another addition to the chorus of frustration with Python 2.7...at least v0.11 should be the last release to support it?)

[dopplershift]

Ok, so there's a work-around for the Python 2.7 issue by adding ImportError to the list of exceptions we catch when trying to tell pint to setup its matplotlib handling. I've added that commit to this PR so we can test if this fixes it. This PR will need a rebase onto latest master to fix the conflict with in setup.py.