Performance variability boxplot computation

[jarusified]

This PR implements a high-level API class, called Boxplot to compute the performance variability of call sites. The Boxplot API computes the quartiles, outliers, and several statistics from a GraphFrame. The results of the computation (indexed by callsites) are dumped as GraphFrames (as members (i.e.,self.gf). The API also exposes to_json method to dump the results in JSON, which can be used to visualize as boxplot vis in #394. The boxplot API takes the following arguments,

• multi_index_gf - Multi-indexed GraphFrame (required),
• drop_index - The index to drop in the ht.GraphFrame.dataframe to compute the variability (e.g., rank, dataset) (optional).
• metrics - list of inclusive/exclusive metrics (optional) [default = inc_metrics + exc_metrics]

bp = BoxPlot(multi_index_gf=gf, drop_index="rank", metrics=["time"])

# The computed GraphFrames can be dumped as below.
print(bp.gf)