Let macro function takes the whole form and the environment

[kchanqvq]

This fixes #517 and is essential towards complete CL. It does introduce some code churn so I'd like some eyes on it!

This introduces parse-macro (from CLtL2) and use it to bootstrap defmacro, instead of using destructuring-bind directly. Currently !parse-macro just has additional &whole handling. In the future we might add &environment as well, which should then make macro lambda list support complete.

[kchanqvq]

There's one more thing that I might want to change. The problems:

1. macroexpand and friends should operate in the null (global) lexical environment when called without env argument, while currently they take current value of *environment*
2. lookup-in-lexenv is not the most efficient and we aren't using the full functionality of push-to-lexenv -- there's at most one definitions in one lexenv frame.

1 can be solved by saving another *global-environment* variable. However I think the cleanest way is to define environment as a list of lexenvs instead (maybe rename to lexenv-frame), and variables functions... can be hash tables. Previously we store a single lexenv struct whose variables functions ... list structures implicitly contain data from parent frames (and we're limited to represent them as list), and the frame structure weren't explicit.

[kchanqvq]

After working on it I think the current environment representation is fine. I won't change it.

Update:

• save global environment in *environment* and use them in macroexpand(-1)
• support &environment parameter

The code we ported (CLOS, loop) seems to have some hacks making up for not supporting &environment before. I'll undo them.

[davazp]

I never looked well how the environments work in CL. 🙂 So hard for me to catch any issue on the PR if there is any.

But let's merge it and see if we find something.