slow bytecode wrappers

[Bike]

During bootstrap, we use bytecode to handle the complex lambda lists of some Lisp functions implemented in C++, like vector-push-extend. These bytecode wrappers are appreciably slower than native-compiled wrappers, so it would be good to replace them.

There is code to do this already in bytecode-adaptor.lisp, but it doesn't work during build. This is because it simply tries to compile every bytecode function, including ones that are not lambda list wrappers. That causes various problems, especially because many of the non-wrapper functions have compile-info in their bytecode wrappers that are not appropriate.

Filing this to stave off forgetting about it while I work on other things.