LLAP: Vector group by operator should use memory per executor

HIVE-15503 treatment has to be applied for vector group by operator as well. Vector group by currently uses MemoryMX bean to get heap usage and heap max memory which will not work for LLAP. Instead it should use memory per executor as upper bound to make flush decision.