[pseudo] Turn glrReduce into a class, reuse storage across calls.

[pseudo] Turn glrReduce into a class, reuse storage across calls.

This is a ~5% speedup, we no longer have to allocate the priority queues and
other collections for each reduction step where we use them.

It's also IMO easier to understand the structure of a class with methods vs a
function with nested lambdas.

Differential Revision: https://reviews.llvm.org/D128301