further performance improvements to MatchSet functionality - factor out MatchSet's State so that we dont' have to keep copying over the state every time we create a new child MatchSet. Also fixes a bug where we could get duplicate match indexes when a new match is created while evaluating an existing match Also, try to be lazy about creating the first MatchSet so that we don't hurt bigtable performance (i.e. when there are no py:matches in play at all)