Can be useful to monitor an ongoing simulation. This cannot be done by
reading the output file, because opening the NetCDF file from two
programs at once may corrupt it and crash either/both programs. Instead,
if `runtime_plots = true` is passed in the input, a .png file with a
bunch of plots of moments and distribution functions is produced at each
output step. The plots are created in serial, and may take some time
(~10s?) for high-resolution simulations.

These can be useful for post-processing.

Also requires updated version of Plots.jl (to be added later if/when
JuliaPlots/Plots.jl#4298 is merged).

`export analyze_and_plot` -> `export analyze_and_plot_data`

Making animations directly with PyPlot.jl and matplotlib is not as neat,
and the output is not consistent with Plots.jl, but it works without
needing the update to Plots.jl in
JuliaPlots/Plots.jl#4298

Needed when using PyCall to import matplotlib.

Using the system Python is the default on Linux, but not macOS (or
Windows). This commit explicitly sets the ENV["PYTHON"] variable so that
all platforms use the system Python. This was needed to get PyCall to
work correctly on macOS - previously Julia's internal Conda-installed
Python had linking errors when using matplotlib.

When creating figures for multiple runs, we need to set the figure size
explicitly for 2d plots (it should get wider when there are more runs to
avoid squashing the figures). This commit changes the settings so that
the size of the plot for each run matches the Plots.jl default size for
a single plot.

After switching between PyPlot interface and Plots interface, the Figure
apparently does not get cleaned up without an explicit call to
Plots.closeall().

Includes some refactoring of functions used to get/plot unnormalized f
so that they can be reused in more ways.