fix: Allow macros to change types on each iteration of a comptime loop

[jfecher]

Description

Problem*

Resolves #6086 (comment)

Summary*

Allows macros to have differing result types on different loop iterations again after it was banned by the unification in #6086.

This is a difficult feature to use since users will still have to work around the type system only type checking loops once. The recommended technique found for this PR is to delay type checks by quoting the relevant function calls that need to be re-typechecked so that when they are unquoted by the interpreter they are forced to be typechecked every loop iteration.

The downside of this is that it requires some knowledge of when to do this. If we go back to the naive approach of comptime_change_type_each_loop_iteration before the quoting:

fn main() {
    comptime {
        for i in 9..11 {
            // Lengths are different on each iteration: "foo9", "foo10"
            let name = f"foo{i}".as_ctstring().as_quoted_str!();
            let hash = from_signature(name);
            assert(hash > 3);
        }
    }
}

fn from_signature<let N: u32>(_signature: str<N>) -> u32 {
    N
}

We get a confusing error now that the unification is gone: "Non-integer array length _" pointing to the body of from_signature. The fix for this is to quote & unquote the from_signature call:

let hash = unquote!(quote { from_signature($name) });

But there's nothing telling users this. Ideally we could warn users when a type changes like this and perhaps have them opt-in to it with a "I know what I'm doing" of some kind.

Additional Context

Documentation*

Check one:

• No documentation needed.
• Documentation included in this PR.
• [For Experimental Features] Documentation to be submitted in a separate PR.

PR Checklist*

• I have tested the changes locally.
• I have formatted the changes with Prettier and/or cargo fmt on default settings.

[github-actions]

Changes to Brillig bytecode sizes

Generated at commit: 0b58c43dedb0aeb1ff3b8e9daf3c79f3de00affb, compared to commit: 4170c55019bd27fd51be8a46637514dfe86de53c

There are no changes in circuit sizes

[jfecher]

As another consequence of this, the macro_result_type test added in the previous PR has also been moved to compile_failure

[asterite]

Looks great! I also tried on the code I have in Aztec-Packages and it works fine.