Improve parallelism of push-based ExecPlanTask for binary operator

[hartig]

The current implementation of PushBasedExecPlanTaskForBinaryOperator (as provided by PR #211) ignores whether op.requiresCompleteChild1InputFirst() is true or false but, instead, simply consumes and pushes the complete child 1 input first (as would be required in case that op.requiresCompleteChild1InputFirst() returns true). To improve parallelism, for the case that op.requiresCompleteChild1InputFirst() returns false, poll both inputs each time until exhaustion before starting to wait for the first input.

[hartig]

The new PushBasedExecPlanTaskForNaryOperator (as added by PR #280) has the same problem.

[hartig]

Commit e728edd adds the code for this (not only for the binary operators but also for the n-ary ones). However, the relevant code paths are currently commented out. Before actually using them, we should do some experiments to better understand their behavior in different cases.