A lot of ambiguities due to too general randsignatures

[thofma]

We are still using this quite heavily in Oscar land, but after playing around with https://github.com/JuliaTesting/Aqua.jl (which helps detecting method ambiguities), I noticed that RandomExtensions.jl leads to a lot of method ambiguities. For example, the methods

rand(X, ::Type{Tuple}, n::Integer)
     @ RandomExtensions ~/.julia/packages/RandomExtensions/qAD6J/src/containers.jl:53
rand(X, t::Type{<:Array}, dims::Integer...)
     @ RandomExtensions ~/.julia/packages/RandomExtensions/qAD6J/src/containers.jl:106

will make the pass-through methods

rand(T::MyAbstractType, v...) = rand(GLOBAL_RNG, T, v...)

ambiguous for certain input types (note that I cannot pin down the type of v directly, since T is abstract and there are many possible combinations).

They are usually fine in practice, but it just looks bad. There are so many, that ignoring them might hide real ambiguities.

[rfourquet]

One problem with these methods in "containers.jl" is that they are doing type piracy, so the plan is to split the bulk of this package into a non-pirating part, which Oscar could be updated to depend on (last time I checked, Oscar was not depending on methods defined in this file).