give IR forms for defining types valid linear structure

[JeffBezanson]

The type-defining IR forms (struct_type etc.) are currently very special: they violate the usual linear structure rules for IR, and evaluate their arguments in a non-standard way intertwined with some global effects. This PR tries to clean that up, removing the special cases and hopefully eventually leading to fixing #33183.

The new type-defining process here is broken into three steps at the IR level:

1. Create a new partially-initialized type object and bind it to a local variable.
2. Evaluate the supertype expression and set it.
3. Evaluate and set field types, fix up partial types.
4. Assign the type globally, doing the special check for valid redefinitions.

Uses only function calls (unfortunately requiring several new builtins though).

fixes #35416

[JeffBezanson]

It seems to me that if we break everything down into simpler forms, JuliaInterpreter can continue to work unmodified, which would certainly be a plus.

[timholy]

Since you asked about JuliaInterpreter, the changes were relatively minor (though not entirely trivial to track down):

• Fix on Julia nightly JuliaDebug/JuliaInterpreter.jl#389 and specifically JuliaDebug/JuliaInterpreter.jl@77660f5
• Fix on nightly JuliaDebug/LoweredCodeUtils.jl#28
• Fix on latest nightly timholy/Revise.jl#473

We'd noticed the former differences in lowering, and even had to handle them specially to make sure a performance optimization didn't trigger errors. So I'm supportive of this change.

[NHDaly]

Hooray! Thanks Jeff! Awesome. :)

[o314]

Currently, eg. in Julia v1.5.2 it still remains a mention to those expr in

    (define (check-top-level e)
      (define (head-to-text h)
        (case h
          ((abstract_type)  "\"abstract type\"")
          ((primitive_type) "\"primitive type\"")
          ((struct_type)    "\"struct\"")
          ((method)         "method definition")
          (else             (string "\"" h "\""))))

@ https://github.com/JuliaLang/julia/blob/v1.5.2/src/julia-syntax.scm#L3786-L3790

I don't know if it is voluntary (for backward compat or something else), or an omission