Allow synthetic control with more than 1 treatment cluster

[Gabrielcidral1]

Currently, performing power analysis with synthetic control are only possible with only 1 treatment cluster and the rest of clusters in control. We would like to allow users to perform any type of split, similar to other analysis methods.

Basically, we need to find the best combination of weights for 1 treatment cluster at a time. So this means we need to run the model sequentially, for every treatment cluster. However, we need to store somewhere which are the treatment cluster because they can't be used as donors.

Also, we need to understand how to calculate the ate and p-values when using more than 1 treatment cluster.

Originally posted by @david26694 in #168 (comment)