optimize `PSet.intersect()`

[hrldcpr]

• [ ] if the passed list is a Set, then iterate over whichever set is smaller, so we get O(mlgn) complexity, with m<n
• [ ] benchmark that this is actually faster

[hrldcpr]

(Also, is there a heuristic for plus-ing the intersection up from empty if it's small, versus minus-ing it down from full (as the current code does) if it's big? Probably nothing consistent…)