[DTable] Negative groupby scaling with threads

[krynju]

Noticed this in my benchmarks
Table shows results for 2, 4, 8 and 16 threads and their comparisons against times from 2-thread config
Other ops mostly scale well

Either the groupby algo is messed up or the scheduler doesn't handle it well

[krynju]

Scaling works fine with processes though
each worker has 4 threads in this setup

4 workers time -> 696.21
8 workers time -> 234.21
speedup -> 2.97