Revisit decision around mapping function selectors to resources vs directly to systems

[alvrs]

Instead of mapping to a resource, the function selector could map direcly to a system function, which would save one sload per call, but add some complexity to upgrading systems. See WorldRegistrationSystem.registerFunctionSelector

[alvrs]

If we wanted to map function selectors directly to system addresses we'd introduce a lot of additional complexity:

• duplicating the access control logic: if a system is not registered as openAccess, we'd have to load the resource selector from storage and perform access control checks based on it.
• duplicating the logic when to call a system via delegatecall or call (which is currently determined based on the resource selector). Probably would have to store it as a flag in the function selector table, so there is not a single source of truth anymore for whether a system is a root system.
• increased complexity for upgrades: instead of just pointing the resource selector to a new address, we now would have to set up a reverse mapping for which function selectors are registered for a given address, and change all those references to the new address too
• increased "conceptual complexity" - instead of once concept for systems (resource selector) we would have two concepts and two sources of truth (ie "a system is a root system if it is registered in the root namespace or if there is a flag in the function selector table saying so")

To avoid all of this complexity I think it's a worthwhile tradeoff to accept a single additional sload and keep pointing function selectors to resource selectors.