Stop unwrapping types while mapping

[timholy]

This started out as just an attempt to get the remaining tests working on Julia 1.11. However, one of the failures was a case where we wanted to preserve a Core.Const wrapper, and as I dug in it became increasingly clear that it was a little risky to alter the type assigned by inference before mapping.

This PR preserves (or, intends to preserve) the exact types assigned by inference. It then becomes the show component's job to strip Core.Const and similar wrapper before displaying types.

@Zentrik, now the remaining failures seem to be in the vscode portions. Any chance you could take a look at this? I'm less familiar with that code.

[timholy]

Oh wait, it ended up being really simple. The question is whether one thinks 0::Core.Const(0) is a regression compared to 0::Int?

[Zentrik]

Seems like an improvement to me given it's much more specific than Int.

[timholy]

Cool. I'll look into that "exhaustive.jl" failure, and if/when fixed we should have TypedSyntax tests passing on 1.10 and 1.11 (but not nightly).

Given that inference is allowed to change between releases, I don't think any of these changes count as breaking? (Cthulhu is also the only dependent of TypedSyntax listed on juliahub.)

[timholy]

Let's see if CodeTracking PR130 fixed the exhaustive.jl test

[timholy]

TypedSyntax is now passing on 1.11. Since this is a fairly significant change and I'm not 100% confident in our test coverage (as good as it is), I'll play with it interactively a little before merging.

[timholy]

This seems to be working reasonably well and without egregious mistakes, so let's merge this.