Create DiagonalTensorMap from SectorVector

[Yue-Zhengyuan]

The new behavior of svd_vals to return a SectorVector instead of DiagonalTensorMap has broken PEPSKit in (at least) the following places (see QuantumKitHub/PEPSKit.jl#314):

• CTMRG convergence check calc_convergence calls _singular_value_distance((S₁, S₂)), which assumes both S₁, S₂ are DiagonalTensorMaps with domain V₁, V₂. It needs to use the space information to project both into the smaller space.
• BPEnv to SUWeight conversion, which only needs the singular values of the product of square roots of two messages on the same bond.

The easiest way to solve these is to add a new constructor of DiagonalTensorMap directly from SectorVectors, from which the (non-dual) domain can also be inferred.

[Yue-Zhengyuan]

A simple implementation of the constructor:

function DiagonalTensorMap(s::SectorVector)
    S = sectortype(s)
    V = Vect[S](c => length(b) for (c, b) in blocks(s))
    return DiagonalTensorMap(s.data, V)
end