Use matplotlib object-oriented API for plotting

[israelmcmc]

There are multiple functions in cosipy using pyplot directly e.g. plt.plot() --i.e. the stateful interface. This is OK for an user-defined or user-facing script, but not in a library, since it makes it hard to control the flow and can conflict with the user customization. See for example the issue described in #481.

Instead, all methods generating a plot should take as input a matplotlib Axes, which will be the one use to produce the plot. The user would do seomething like

# Create Axes
fig,ax = plt.subplot()

# Plot using cosipy
foo.plot(ax)

# Customize
ax.set_xlim(5,10)

Internally, the plotting function should look like:

def plot(ax = None):

    if ax is None:
        # Create Axes it not provided
        fig,ax = plt.subplots()

    x,y = compute_data() 
    ax.plot(x,y)

    return ax

See also histpy and mhealpy for more examples.

[GallegoSav]

@israelmcmc This does not concern plots that are made by the plot() and draw function from histpy/mhealpy right ?

Like here for example

[israelmcmc]

@GallegoSav That example is halfway of fixing the problem. Indeed using histpy/mhealpy returns the ax object, but we would also need to:

1. make inject_model accept an ax parameter
2. Pass ax to histpy's draw
3. Return ax in inject_model

Furthermore, it would be much better if the plotting methods would only do plotting and the plotting were done only by plotting methods. That is, inject_model shouldn't be doing any plotting, not even with an optional parameter, only the calculation. Instead, we could have a plot_injected_model(ax, model) function that the user can call at a later step.