Allow more circular type definitions

[martinholters]

Checking jl_types_equal against a potential wrapper when trying to normalize type parameters can result in an infinite recursion for circular type definitions. E.g.:

julia> struct T22624{A,B,C}; v::Vector{T22624{Int64,A}}; end # ok and tested for

julia> struct T22624b{A,B,C}; v::Vector{T22624b{Integer,A}}; end
ERROR: StackOverflowError:

Notably, the former only works due to an invalid fast-path, making it also recurse infinitely in #25796 and #31807 (without appropriate countermeasures). The culprit is that jl_types_equal needs to rename_unionall here, which tries to re-instantiate the same type (with another typevar), which comes to same test again. The obvious counter-measure is to avoid jl_types_equal and instead use a heuristic that may miss opportunities for normalization, but will not lead to recursion. As a starter, I'm using obviously_egal here, which fixes above issue, but may be too restrictive, e.g.

julia> Ref{Array{T,N} where {N,T}}
Ref{Array} # master
Ref{Array{T,N} where T where N} # using obviously_egal

Tests still pass, but we probably miss some optimization opportunities this way. Though I have no idea how much impact this actually has, so it might even be ok like this.

Discussion of this started in #25796, but this issue is somewhat independent, so for the sake of more focused discussions, I'm opening this draft PR.

[fingolfin]

The example in the PR description has been working since at least Julia 1.2. So I think it is OK to close this PR now