inference: Remove special case for typename

[Keno]

This was introduced in [1] to improve constant propagation. These days this is handled via concrete eval and the special case is no longer required.

[1] 0292c42

[Keno]

There's a remaining question here of whether we're losing any precision for the ::Type case and if so whether that matters.

[Keno]

We do lose this one:

julia> f(x) = Base.typename(x)
f (generic function with 1 method)

Before:

1-element Vector{Any}:
 CodeInfo(
1 ─     return $(QuoteNode(typename(Array)))
) => Core.TypeName

After:

julia> code_typed(f, Tuple{Type{Union{Vector{Int64}, Vector{Float64}}}})
1-element Vector{Any}:
 CodeInfo(
1 ─ %1 = invoke Base.typename(x::Union)::Any
└──      return %1
) => Any

Let me think about whether there's something that can be tweaked here with effect annotations.

[Keno]

If not, the easy answer is just to put this one in Core and compare it by value, which doesn't have the same problems.

[Keno]

If not, the easy answer is just to put this one in Core and compare it by value, which doesn't have the same problems.

I think I'll probably just do this. The key extra bit of information that the compiler does not know is that checking that they're all equal retroactively makes taking union components out legal, but I don't see a good generic way to annotate that.

[aviatesk]

This PR is going to be replaced by #55289 .