require_relative implementation.

…is saved

somewhere!

As systemic insulating design choice all identifiers are forced to construct
an eager RubySymbol as constructor argument to the AST (note: parsing and IR
APIs definitely are not backwards compat with 9.1).  This should future-
proof things.  By forcing the AST to use symbols it means we cannot make idents
which are not eager in making a symbol table entry.  This in turn means later
on when we work with the raw string equivalent of these idents....we know there
will be a symbol table entry.  As an aside MRI is eager on all idents/symbols
so we are being a little more like them (which is not semantically important
but who knows?).

Constants, ivars, gvar, cvars all made to use raw strings.  This still has some
fallout to deal with. For example const_get? will use the string and assume it
can charAt but it cannot since it is now a raw string.  In cases like this
we need to ask for properly encoded bytelist and perform same logic but in a
mbc friendly way. This will likely be the next commit(s).

Tons of error reporting corrections (with tons more needed).  Now that all
identifiers are raw strings they look weird when they contain mbcs.  So I added
a buildString which makes a RubyString which will do proper m17n stuff to make
a byte coherent string.  I did many of these when I see them fail in the test
suite but the pure lack of coverage in this regard is staggering.  My hope to
limit an endless stream of reports is to lean on all errors which refer to
idents using the same helper method as a visual cue that it will be properly
handled.  Both this and the symbols in AST constructors are meant to help us
at a glance not make an error or know why we have an error at a glance.

All-in-all this last dump has had me work through many issues and I am still
pretty happy.  Making older extensions continue to work has added a tough
dimension but so far so good :)