Fix conj to use deepcopy instead of copy

[jofrevalles]

Summary

In this PR, we change Base.conj function to use deepcopy instead of copy to prevent errors when manipulating tensor networks within an @unsafe_region macro block.

The regular copy leads to problem since it calls TensorNetworks(tensors(tn)), which throws an error since there are inconsistent indices. Before the recent commit a241967 that refactored the conj and adjoint functions, we also had a deepcopy inside conj which prevent this problem.

Example

With this PR, we can do something like the following code without any problem:

julia> using Tenet

julia> tn = TensorNetwork([
           Tensor(ones(2, 2), [:a, :b]),
           Tensor(ones(2, 2), [:b, :c])
           ])
TensorNetwork (#tensors=2, #inds=3)

julia> Tenet.@unsafe_region tn begin
           c = Tensor(ones(3, 2), [:c, :d])

           push!(tn, c)
           conj(tn)
           pop!(tn, c)
       end
3×2 Tensor{Float64, 2, Matrix{Float64}}:
 1.0  1.0
 1.0  1.0
 1.0  1.0

[jofrevalles]

@mofeing Okay I got another idea, we can keep the copy but add modify it like this:

function Base.copy(tn::TensorNetwork)
    if is_unsafe_region[]
        new_tn = TensorNetwork([])
        for tensor in tensors(tn)
            push!(new_tn, copy(tensor))
        end

        return new_tn
    else
        return TensorNetwork(tensors(tn))
    end
end

Since push! accounts for the unsafe_region. Or we can directly add that conditional inside the TensorNetwork constructor.

[mofeing]

Okay, I like this idea more. The one about adding the check inside TensorNetwork constructor.

[jofrevalles]

Okay, I like this idea more. The one about adding the check inside TensorNetwork constructor.

@mofeing If we do this, then we should agree to change the macro to my second option described in issue #230. For me, it makes sense.

[mofeing]

Okay, I like the idea but we have to discuss consequences (e.g. we might have problems when multithreading). Add it to the agenda for today's meeting please.