Replace hard-coded truncate! threshold with the epsilon value of the chain's eltype

[jofrevalles]

Summary

This PR modifies the previously hard-coded threshold inside the truncate! function, which was fixed to 1e-16, to the proper epsilon value of the chain's eltype, making it more flexible and general.

[jofrevalles]

Now everything seems to work fine:

julia> using Tenet

julia> A = Tensor(rand(Float32, 2, 2), (:i, :j))
2×2 Tensor{Float32, 2, Matrix{Float32}}:
 0.252758  0.719538
 0.417894  0.9425

julia> B = Tensor(rand(Float32, 2, 2), (:k, :j))
2×2 Tensor{Float32, 2, Matrix{Float32}}:
 0.802961  0.219748
 0.143793  0.835135

julia> tn = TensorNetwork([A, B])
TensorNetwork (#tensors=2, #inds=3)

julia> Tenet.wrap_eps(eltype(tn))
1.1920929f-7

julia> A = Tensor(rand(ComplexF64, 2, 2), (:i, :j))
2×2 Tensor{ComplexF64, 2, Matrix{ComplexF64}}:
 0.500145+0.230456im   0.644049+0.482663im
 0.599711+0.0412583im    0.7519+0.889049im

julia> B = Tensor(rand(ComplexF64, 2, 2), (:k, :j))
2×2 Tensor{ComplexF64, 2, Matrix{ComplexF64}}:
 0.0460109+0.810682im   0.55873+0.698605im
   0.58918+0.183138im  0.733106+0.15675im

julia> tn = TensorNetwork([A, B])
TensorNetwork (#tensors=2, #inds=3)

julia> Tenet.wrap_eps(eltype(tn))
2.220446049250313e-16