use specialize kernels in rolling_groupby aggregation ~10x perf gain (window of 100 elements)

[ritchie46]

If we have group slices and they overlap, we can use the fast rolling_<agg> kernels.

Benchmark on a rolling window with 10_000 elements and a window size of 100:

Gnuplot not found, using plotters backend
bench_1                 time:   [107.76 us 108.04 us 108.33 us]                    
                        change: [-91.314% -91.244% -91.167%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 8 outliers among 100 measurements (8.00%)
  1 (1.00%) low mild
  5 (5.00%) high mild
  2 (2.00%) high severe