POC/REF: remove SeriesBinGrouper, SeriesGrouper

[jbrockmendel]

Maintenance of libreduction is a PITA, so I want to see how big a performance hit we take if we rip it out.

I get a test failure locally because apparently something in resample behaves differently if we go through the python path, which itself should be considered a bug. cc @mroeschke any ideas how to fix the failing test/

[jbrockmendel]

asv run with -b groupby:

       before           after         ratio
     <master>         <no-_aggregate_series_fast>
+         135±3ms       1.75±0.03s    12.94  groupby.MultiColumn.time_lambda_sum
+        71.6±1ms         881±10ms    12.31  groupby.MultiColumn.time_col_select_lambda_sum
+      14.0±0.3ms       15.7±0.3ms     1.12  groupby.Nth.time_frame_nth('float64')

[jbrockmendel]

Closing to clear the queue, will revisit. Still hoping to @mroeschke can take a look at the failing test case, since that suggests an extant bug hidden somewhere.