Refactor Summarizers

[jc-healy]

SummarizerDisplayWidgets by default are panel widgets that are instantiated by summarizers which return dataframes.
HvplotDisplayWidget
DataTableDisplayWidget
Eats a dataframe.
Promote summarizers.dataframe to .summarizers build a smattering of dataframe returning summarizers.
ContrastiveWordImportanceSummarizer. List of text in the constructor -> CountVectorizer -> NgramVectorize(1,4) ->
CoherentWordImportanceSummarizer. USE embedding of lists of text. TextEncoder=None -> use. Otherwise allow a model to passed in. Models have an encode method that take a list of strings. Tokenize and get keywords out.
FeatureImportanceSummarizer
ValueCountSummarizer

Docs page describing a bunch of ways to use them with lots of hvplot examples.

Summarizers which return things other than dataframes will be moved to a summarizers.custom and will eventually be paired with custom DisplayWidgets. For now hvplotDisplayWidget should provide enough flexibility for most of our use cases.

SelectorWidget (bidirectional linking a means of selecting data that is independent of the scatterplot pane. This is perhaps not a class for now. We'll do them as a set of one offs. This it moves the TimeSeriesSelection out of the summarizers. A read the docs page describing these as a coherent class of first class citizens and what it takes to build one would be great.
TimeSeriesSelectorWidget
Search is a top level selector Widget
Legends is a top level selector Widget
In some sense BokehPlotPane is a top level selector Widget.