Smart caching for input->AST conversion in the engine.

[charithe]

One of the hotspots in the engine is the toAST function which converts the incoming request into a Rego AST for processing. With real life usage, it is very likely that the incoming requests will contain the same principal and resource fields (but in different combinations). With some clever caching, we may be able to reduce the amount of work we do on every request.

Another avenue of exploration is to attempt lazy-loading using the Resolver interface. Instead of converting the whole input to AST every time, the Resolver could help us selectively convert portions of the input as needed.

[charithe]

Superseded by #354