Currently, the causal impact plots have HDI's which represent the model expectation.

This was done intentionally, but we should do the following:

• Add a kwarg to allow used to decide whether to plot the model expectation or the posterior predictive.
• Explain this kwarg in the appropriate docstrings
• Make sure that this is explicit, for example by explaining this in the example notebooks, or by a text output when the plot method is called, or by adding a text annotation to the plot.
• It would be good to (by default) add a bit of text to the figure to explain what the HDI corresponds to. Maybe small text size at the bottom of the key (often 3-panel) plots.