Fix builtin_tfunction for Core._expr

[martinholters]

This had no effect before, as Expr is not a Builtin, but Core._expr is.

Before:

julia> code_typed(Expr)
1-element Vector{Any}:
 CodeInfo(
1 ─ %1 = Core._apply_iterate(Core.iterate, Core._expr, args)::Any
└──      return %1
) => Any

After:

julia> code_typed(Expr)
1-element Vector{Any}:
 CodeInfo(
1 ─ %1 = Core._apply_iterate(Core.iterate, Core._expr, args)::Expr
└──      return %1
) => Expr

[martinholters]

(Some interesting(?) archeology: I think this was missed in bec065f (#8712), so this is actually a fix to a regression ... introduced between 0.3 and 0.4 😄)

[JeffBezanson]

The if isva; return Any check in there is also pretty pessimistic, e.g.

julia> f(a...) = ===(a...)

julia> code_typed(f)
1-element Vector{Any}:
 CodeInfo(
1 ─ %1 = Core._apply_iterate(Base.iterate, Main.:(===), a)::Any
└──      return %1
) => Any

[martinholters]

Yes, I had considered going further here, but this change was so obvious and seemed sufficiently beneficial considering Expr(:head, args...)-like constructs that I wanted to get it in before getting lost in some larger rework. But I now do have this on my mind...