fix #32488, type intersection regression

[JeffBezanson]

Issue was caused by 74305bf, so this reverts part of the new logic from there.

There is a loss of precision in one test case, which I easily consider acceptable.

[tkluck]

This seems to cause a new method ambiguity for me (copied below). I used git bisect to point me to this PR.

The ambiguity looks like a bug, because one of the alternatives doesn't actually apply to the given types. Moreover, using methods shows only a single applicable method.

I haven't succeeded in isolating a MWE yet. The below comes from using the master version of https://github.com/tkluck/PolynomialRings.jl :

julia> using PolynomialRings

julia> R = @ring Int[x]
Int64[x]

julia> convert(R, :x)
ERROR: MethodError: convert(::Type{Int64[x]}, ::Symbol) is ambiguous. Candidates:
  convert(::Type{P}, a::C) where {C, P<:(<undisplayable> where MonomVector where #s23<:PolynomialRings.Monomials.AbstractMonomial{Order} where Order)} in PolynomialRings.Conversions at /home/tkluck/.julia/dev/PolynomialRings/src/PolynomialRings/Conversions.jl:51
  convert(::Type{P}, x::Symbol) where P<:Polynomial in PolynomialRings.EntryPoints at /home/tkluck/.julia/dev/PolynomialRings/src/EntryPoints.jl:25
Possible fix, define
  convert(::Type{P<:(<undisplayable> where C)}, ::C<:Symbol)
Stacktrace:
 [1] top-level scope at REPL[3]:1

julia> methods(convert, Tuple{Type{R}, Symbol})
# 1 method for generic function "convert":
[1] convert(::Type{P}, x::Symbol) where P<:Polynomial in PolynomialRings.EntryPoints at /home/tkluck/.julia/dev/PolynomialRings/src/EntryPoints.jl:25

(to avoid confusion: it says <undisplayable> because I try to pretty-print some types in some cases.)

Does this ring a bell?

[tkluck]

^^ The issue I'm reporting in my previous comment seems to have been fixed as of a18ab97. Thanks!