Improve performance of merging lists of policyEvaluators returned by Trie

For a multi-level resource, Ranger policy evaluation, to some extent, depends on how the policies are written, and how the a set of likely matching policies is evaluated. This is traced back to sub-optimal algorithm used for computing intersection of sets of matching policies for each resource level in the accessed resource. Improving this algorithm is required to a. improve performance of policy evaluation, and b. improve predictability of performance of policy evaluation.