Avoid lambda allocation in AtomicInteger max update in MutablePQVectors

[michaeljmarshall]

This is a minor optimization to reduce object allocations on the hot path when we're adding vectors to MutablePQVectors. The object allocation happens when we have a capturing lambda, which is different on every call to this method. Instead, we can use the accumulateAndGet method, which takes the value we want to offer and then takes a non-capturing lambda Math::max, which means we won't allocate an object per call.