Enforce a "Spill Maximum" per query

Currently one query that spills excessively to disk can use up all cluster space and cause other queries/workloads to fail.  I understand that a global maximum would be difficult to track, but some limit per mapper/reducer would help keep the spill under control.