Delimit continuations via enums instead of functions

[alaviss]

The idea is to use state enums to delimit continuations, the benefits are simple:

• One runner function, so there is less functions and linkers can potentially inline calls.
• The states can be portable between different instances of the program as there are no pointers that might change between JS and C.
• Allow for us to define how yielding works in CPS. Right now a CPS function will yield on every split, including the ones that the programmer doesn't care about. This may cause performance issues as yields involves going in and out of functions.

This is a big change and how to represent the enum is still an open question (ie. if its in Continuation, then it must be a generic type like int).

The priority on this is undecided, might be after generics.

[disruptek]

My interpretation of #183 was the simple post-CPS generation of this stuff; insert a type, install it in the continuation, etc.

I think we can be flexible with respect to the type, but I don't want to bury it in the environment. Even with the inspector, it's just a PITA and we paid good money for the Continuation abstraction -- I want to use it.

I'd also want to rethink how hooks work and either put them into the same state machine semantic or develop a new way to specify/implement them.

I know hooks are not a popular feature, but if they are too expensive to maintain, I think that's the fault of the implementor and not the fact that they are not valuable enough to justify an implementation. So let's try to fix that or convince each other that they need to be removed fully.