Add parallelism control

[je-ik]

After removing explicit partitioning, we have currently no explicit control over the parallelism of executing operators. This affects both batch and stream. There must be a way to give a hint to the translator that certain operation should be parallelized more or less than the input. Options are:

• add a method to set parallelism of operator on executor - e.g.

    Executor executor = ...;
    executor.withParallelism("OPERATOR_NAME", 100).submit(flow);

• add downstream parallelism hint to shuffle operators, e.g.

   ReduceByKey.of(...)
       .keyBy(...)
       ....
       .withHint(Parallelism.of(100));

• some other option?

[dmvk]

I think we should never set explicit parallelism, instead we should hint operator with the percentual estimate of increase / decrease in data size, so we can decide parallelism based on the input data.