Add multiple ways of automatic trend changepoint detection, for instance BEAST

Much of the merit of a decomposition model comes from successfully distinguishing trend changepoints from seasonal, or residual. Most people don't/barely look at changepoint detection, and if they do, unless they have a proper clue of what the trend is doing and change the parameters for it, the resulting decomposition will be wrong. This affects their forecasts and incurs in an overall mismatch between different components of the TS.
Most users might overlook this, and have the model use the standard 0.005 scale, which might very much not adjust to their needs.
I suggest to implement new/other ways to identify trend changepoints. For instance Bayesian Ensemble Algorithm (BEAST), which already has a python implementation: https://github.com/zhaokg/Rbeast#python

This will improve the overall implementation of Prophet.