Resolve core attribute arguments, runtime setters, safely

[rurban]

Some builtin attributes will need to resolve symbolnames :native($libname) at run-time, not only constants at parse-time. Users might want the same, as for users the import call gets the string also verbatim, which has to call eval on it.

This was why Attribute::Handlers was added, but this is bad practice, unsafe and should not be used. Esp. not for core builtins, and there's still no way set CV properties at run-time, such as native <=> dl_load_file(CvFFILIB(cv)) or CvXSUB = dl_find_symbol(:symbol($name)).
So we need a way to set those CV properties.
For strings maybe the UTF8 status or FLAGS. Or ENCODING, if we add that.

Originally perl added an op, such as prototype(cvref|name). This does not scale for more properties. It would need new keywords native and symbol for the ffi. No.

apply_attrs() creates the following: attributes->import(stash, sv, args), which we should use
to do attributes->import(__PACKAGE__, \&cv, "native", $libname);
attributes.pm/xsutils would then be extended to understand it. We only need to defer the load_module call to run-time.

We could also add a MODIFY_CODE_ATTRIBUTES handler to core to catch these new attributes.
But these work only on package names, not types, and we would only add it for main::. Which would make it hard for other packages. So it's a very bad idea.

Better would be generic introspection, as in a proper language or perl6. i.e. treat a CVREF as object with access to all fields. autobox alike.
The setter would be a lvalue method then: \&mysql_ffi_fetch->NATIVE = "mysqlclient.6.so";
Use with via the cvref or via the GV slot:
*mysql_ffi_fetch{CODE}->NATIVE = "mysqlclient.6.so";
*mysql_ffi_fetch{CODE}->SYMBOL = "mysql_fetch";
The 1st cvref method syntax-errors with Can't modify reference to subroutine entry in list assignment, the 2nd GV method errors with Can't call method "NATIVE" on unblessed reference at run-time.

We need it only for the internal setter at run-time, it doesn't even need to be parsed. Though it would be beneficial to do so, if not as keyword/op.

This will be fixed with the ffi merge.