Problems parsing using `:openml` or `:auto` options

julia> df = OpenML.load(470, parser=:openml)
ERROR: MethodError: copyto!(::CategoricalVector{String, UInt32, String, CategoricalValue{String, UInt32}, Union{}}, ::ChainedVector{String, Vector{String}}) is ambiguous. Candidates:
  copyto!(dest::Union{SubArray{var"#s5", N, var"#s4", I, L} where {var"#s5", var"#s4"<:(CategoricalArray{T, var"#s3", R, V, C, U} where {var"#s3", V, C, U}), I, L}, CategoricalArray{T, N, R, V, C, U} where {V, C, U}} where {T, N, R<:Integer}, src::AbstractArray) in CategoricalArrays at /Users/anthony/.julia/packages/CategoricalArrays/rDwMt/src/array.jl:615
  copyto!(dest::AbstractVector{T} where T, src::ChainedVector) in SentinelArrays at /Users/anthony/.julia/packages/SentinelArrays/1hMOA/src/chainedvector.jl:326
Possible fix, define
  copyto!(::Union{SubArray{T, 1, var"#s4", I, L} where {T, T1, R<:Integer, var"#s4"<:(CategoricalArray{T1, var"#s3", R, V, C, U} where {var"#s3", V, C, U}), I, L}, CategoricalVector{T, R, V, C, U} where {C, U, T, R<:Integer, V}}, ::ChainedVector)

This appears to be a CategoricalArrays issue.

However, if highlights the need for more testing.

cc @jbrea (just FYI)

Start by reproducing the OpenML.load examples with parser=:openml and parser=:auto, then read the related CategoricalArrays issue and MLJOpenML issue #16 on testing. Done means the affected parsing options no longer produce the reported copyto! ambiguity and regression coverage is added.