Evaluate parser-level instead of rule-level macro expansion

[sirthias]

Currently we employ a per-rule macro expansion approach which has the benefit of being simpler and more focused than the alternative of defining a complete (or partial) parser in a single macro, similar to what is done in https://github.com/begeric/FastParsers.
It also has the benefit that the user directly controls all aspects of the parser class (inheritance, additional members, etc.) without a macro potentially getting in the way.

The disadvantage is that a per-rule macro expansion has only local knowledge and cannot "look into" other rule definitions, which prohibits catching things like #117 or performing cross-rule optimizations.

We need to evaluate whether parser-level macro expansion has a positive cost/benefit when compared with the current approach.

Also, it might be possible to publish the inner rule structure to the outside even with the per-rule macro approach by moving to structural types as rule macro results, thus potentially combining the benefits of both approaches.