Avoid top-level import of pyplot for benefit of upstream packages

[taldcroft]

This is a possible fix for problems with ORviewer (vis a vis the xija import) where we want to avoid a pyplot import for tools that may just be using a non-interactive backend.

This package has no tests, so testing is not so trivial.

Functional testing

Matplotlib backend not set on import

In [1]: import matplotlib

In [2]: matplotlib.get_backend()
Out[2]: 'MacOSX'

In [3]: from Ska.Matplotlib import *

In [4]: matplotlib.get_backend()  # Default backend not changed
Out[4]: 'MacOSX'

In [5]: matplotlib.use('Agg')  # Can still set backend

In [6]: import matplotlib.pyplot

In [7]: matplotlib.use('MacOSX')  # After pyplot import cannot set backend
/Users/aldcroft/miniconda3/envs/ska3/lib/python3.6/site-packages/matplotlib/__init__.py:1405: UserWarning: 
This call to matplotlib.use() has no effect because the backend has already
been chosen; matplotlib.use() must be called *before* pylab, matplotlib.pyplot,
or matplotlib.backends is imported for the first time.

  warnings.warn(_use_error_msg)

Plot functions work

In [9]: plot_cxctime([0, 100], [0, 100])
Out[9]: 
(((matplotlib.dates.SecondLocator,
   {'bysecond': (0, 15, 30, 45)},
   '%H:%M:%S',
   matplotlib.dates.SecondLocator,
   {'bysecond': [0, 5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55]}),),
 <matplotlib.figure.Figure at 0x7fb0189b4d68>,
 <matplotlib.axes._subplots.AxesSubplot at 0x7fb068d05c88>)

In [10]: matplotlib.pyplot.savefig('junk.png')

In [11]: from Ska.Matplotlib.lineid_plot import *

In [12]: plot_line_ids?

In [13]: plot_line_ids([0, 10], [0, 10], [5], ['label'])
Out[13]: 
(<matplotlib.figure.Figure at 0x7fb0191212b0>,
 <matplotlib.axes._axes.Axes at 0x7fb0191210f0>)

In [14]: matplotlib.pyplot.savefig('lines.png')

In [15]: !open lines.png

In [16]: !open junk.png

[jeanconn]

I like this (and considered it as well) but wasn't sure we wanted it for the with-proseco-4.4 release. For that, I thought a change in sparkles to switch to the non-interactive backend if already defined (or to just set to backend-independend non-interactive plotting) might be considered to have smaller scope.

[taldcroft]

I thought the problem is not sparkles but xija. I thought it wasn't possible to change backend once it has been (really) defined, but that might be wrong. By "really", I mean matplotlib.use(backend) doesn't seem to actually do anything irreversible, but once you import matplotlib.pyplot or matplotlib.backends, it seems that the backend is locked in. Again I could be wrong.

[taldcroft]

I've done some functional testing and re-reviewed this. I think it's OK now.