Remove overloads on similar

[adrhill]

These are currently only required for NNlib compatibility with convolutional layers, as the following error is otherwise thrown:

 TaskFailedException
  
      nested task error: UndefRefError: access to undefined reference
      Stacktrace:
       [1] getindex
         @ ./essentials.jl:14 [inlined]
       [2] getindex
         @ ./subarray.jl:290 [inlined]
       [3] conv_direct!(y::SubArray{Dual{Float64, GradientTracer{DuplicateVector{Int64}}}, 5, Array{Dual{Float64, GradientTracer{DuplicateVector{Int64}}}, 5}, Tuple{Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, UnitRange{Int64}, Base.Slice{Base.OneTo{Int64}}}, false}, x::SubArray{Dual{Float64, GradientTracer{DuplicateVector{Int64}}}, 5, Array{Dual{Float64, GradientTracer{DuplicateVector{Int64}}}, 5}, Tuple{Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, UnitRange{Int64}, Base.Slice{Base.OneTo{Int64}}}, false}, w::SubArray{Float64, 5, Array{Float64, 5}, Tuple{Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, UnitRange{Int64}}, true}, cdims::NNlib.DenseConvDims{3, 3, 3, 6, 3}, ::Val{(2, 2, 1)}, ::Val{1}, ::Val{(0, 0, 0, 0, 0, 0)}, ::Val{(1, 1, 1)}, ::Val{(1, 1, 1)}, fk::Val{false}; alpha::Dual{Float64, GradientTracer{DuplicateVector{Int64}}}, beta::Bool)
         @ NNlib ~/.julia/packages/NNlib/g1Wwt/src/impl/conv_direct.jl:111
       [4] conv_direct!(y::SubArray{Dual{Float64, GradientTracer{DuplicateVector{Int64}}}, 5, Array{Dual{Float64, GradientTracer{DuplicateVector{Int64}}}, 5}, Tuple{Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, UnitRange{Int64}, Base.Slice{Base.OneTo{Int64}}}, false}, x::SubArray{Dual{Float64, GradientTracer{DuplicateVector{Int64}}}, 5, Array{Dual{Float64, GradientTracer{DuplicateVector{Int64}}}, 5}, Tuple{Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, UnitRange{Int64}, Base.Slice{Base.OneTo{Int64}}}, false}, w::SubArray{Float64, 5, Array{Float64, 5}, Tuple{Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, UnitRange{Int64}}, true}, cdims::NNlib.DenseConvDims{3, 3, 3, 6, 3}; alpha::Dual{Float64, GradientTracer{DuplicateVector{Int64}}}, beta::Bool)
         @ NNlib ~/.julia/packages/NNlib/g1Wwt/src/impl/conv_direct.jl:50
       [5] conv_direct!
         @ ~/.julia/packages/NNlib/g1Wwt/src/impl/conv_direct.jl:47 [inlined]
       [6] (::NNlib.var"#307#311"{@Kwargs{}, NNlib.DenseConvDims{3, 3, 3, 6, 3}, SubArray{Dual{Float64, GradientTracer{DuplicateVector{Int64}}}, 5, Array{Dual{Float64, GradientTracer{DuplicateVector{Int64}}}, 5}, Tuple{Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, UnitRange{Int64}, Base.Slice{Base.OneTo{Int64}}}, false}, SubArray{Float64, 5, Array{Float64, 5}, Tuple{Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, UnitRange{Int64}}, true}, SubArray{Dual{Float64, GradientTracer{DuplicateVector{Int64}}}, 5, Array{Dual{Float64, GradientTracer{DuplicateVector{Int64}}}, 5}, Tuple{Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, UnitRange{Int64}, Base.Slice{Base.OneTo{Int64}}}, false}})()
         @ NNlib ~/.julia/packages/NNlib/g1Wwt/src/conv.jl:209

However, this does look like a workaround for unusual behavior of NNlib, since the following line most likely shouldn't try to access undef variables:
https://github.com/FluxML/NNlib.jl/blob/381a41f2b912eb924f1a6c256a9bf49a593e4b67/src/impl/conv_direct.jl#L111

[adrhill]

Related fix in NNlib: FluxML/NNlib.jl#592