Fix introspection macros for getproperty

[Liozou]

Non-@code_ introspection macros currently fail on calls to getproperty with the dot syntax. The reason is that @which should not answer the same thing depending on whether it's a call to getproperty with a module or not, i.e. the two following behaviors are expected:

julia> @which Base.nothing
Core

julia> x = LinearAlgebra.lu(rand(Int, 3, 3));

julia> @which x.factors
getproperty(F::LU{T,var"#s825"} where var"#s825"<:(StridedArray{T, 2} where T), d::Symbol) where T in LinearAlgebra at /home/liozou/julia/usr/share/julia/stdlib/v1.6/LinearAlgebra/src/lu.jl:305

This is something I overlooked in #35522. And currently the second case fails:

julia> @which x.factors
ERROR: expected tuple type
Stacktrace:
 [1] error(::String) at ./error.jl:33
 [2] to_tuple_type(::Any) at ./reflection.jl:764
 [3] which(::Any, ::Any) at ./reflection.jl:1133
 [4] top-level scope at /home/liozou/julia/usr/share/julia/stdlib/v1.6/InteractiveUtils/src/macros.jl:72

This PR fixes that. It unfortunately breaks a test that I had intentionally put in #35522. The reasoning was that the two different calls with the dot syntax were ambiguous, so we have to error in the latter case. But actually the two calls are not ambiguous: we simply have to check whether the left-hand side of the dot is a module or not, which cannot be done at macro expansion but can certainly be done runtime.

Anyway, bottom line is:

• @which now differentiates between SomeModule.x and foo.bar (where foo is not a module) and now works in both cases (instead of only the first)
• @less, @edit, @functionloc don't fail anymore on calls to getproperty with the dot syntax.

[Liozou]

On second thought, I don't know if that is the best way to go. The part with @with seems unambiguous, but for the other macros it might still be better to error because I don't think someone typing @edit Base.isabstracttype (for example) wants to see the definition of getproperty. I'm leaving this open to discussion for the time being.