Cache the Lucene query instead of re-creating it from the filter's AST

The AST is already cached so the parsing effort is skipped for both param and param-less queries when are executed repeatedly.

A further minor perf improvement can be achieved by caching also the output of the AST to Lucene transformation phase. This can only be done for queries without parameters (because the params would become bound to the Lucene query objects - which are immutable, btw). The Lucene transformation is a low complexity/effort process but even so it is better to skip it entirely when possible.