Add some support for outlier detection and cleaning

[robjhyndman]

Add some of the functionality of tsclean() and tsoutliers from the forecast package.

Related question: https://stackoverflow.com/q/59051260/144157

[mitchelloharawild]

Issue for outliers is here: tidyverts/fable#160
In the first instance, I think outlier cleaning should be done as a combination of outlier detection, removal, and then interpolation. Simpler/automatic methods can be added later (much like automatic model selection à la forecast::forecast.ts()), where the results are to improve over time as the model selection algorithm improves.

[mitchelloharawild]

Moved to {fabletools} as this functionality will first be implemented with the generics outliers() and interpolate(), where some intermediate function is used to replace outliers with missing values.

A higher level utility function to do these three steps (like tsclean()) can be introduced later, once the necessary parameters and appropriate defaults are better known.

[SeanRichterWalsh]

Hi. What is the progress status on this? I use the forecast package but am interested in whether I should begin modifying code to fable functions. Thanks.

[mitchelloharawild]

At minimum this functionality requires #137.
I have a somewhat clear idea on the design of this feature (using distributional model fits), so basic/default model-based outlier detection would be a small extension once #137 is implemented.

As for prioritisation: I'm working on forecast reconciliation at the moment, however @robjhyndman is actively thinking about and working on outliers and so I expect outliers would be the next priority.