Apply parallel sum gadget optimization to Prio3Histogram

[divergentdave]

The range check portion of the Prio3Histogram circuit could be rewritten to use the parallel sum gadget. This would result in more wire polynomials, each of smaller degree, smaller proofs ($O(\sqrt{n})$ instead of $O(n)$), faster proving, and larger preparation message shares ($O(\sqrt{n})$ instead of $O(1)$). See #124 (comment).