Right now every dataflow algorithm uses its own worklist implementation.
This is a first step to reduce this duplication. Some upcoming
algorithms such as the lifetime analysis is going to use the factored
out implementations.
Differential Revision: https://reviews.llvm.org/D72380