[NDTensors] split dual into UnitRangeDual and GradedUnitRangeDual

[ogauthe]

This PR aims to simplify the use of GradedAxes.dual, especially to be used in a BlockSparseArray. It proposes to split the UnitRangeDual type into two different types:

• a generic UnitRangeDual{<:AbstractUnitRange} <: AbstractUnitRange
• a labelled GradedUnitRangeDual{<:LabelledInteger,<:AbstractGradedUnitRange} <: AbstractGradedUnitRange

This should simplify slicing a BlockSparseArray with dual axes and allows to preserve labels or take their dual.

This branch is built on top of #1363 and should only be merged once #1363 is completed. It does not make any change in Sectors.

[ogauthe]

Currently, this branche solves:

• preserving labels in blocklengths(dual(...))

g = gradedrange([U1(1) => 2, U1(2) => 2])
@test GradedAxes.blocklengths(dual(g)) isa Vector{<:LabelledInteger}  #    # broken in main and in non-abelian_fusion

• preserving GradedUnitRange behavior when slicing:

blocklengths(dual(gradedrange([U1(0) => 2, U1(1) => 2])[1:3])) isa Vector  # broken in main and in non-abelian_fusion

• manipulating arrays with DualGradedUnitRange

r = gradedrange([U1(0) => 2, U1(1) => 2])[1:3]
a = BlockSparseArray{Float64}(r, r)
b = BlockSparseArray{Float64}(dual(r),dual(r))
@test a[:, :] isa BlockSparseArray   # broken in main and in non-abelian_fusion
@test b[:, :] isa BlockSparseArray   # broken in main and in non-abelian_fusion

Meaning one can create and manipulate BlockSparseArray with GradedUnitRange or its dual (not just BlockedOneTo). However axes(a[:,:]) are still Base.IdentityUnitRange(GradedUnitRangeDual{...}), which creates other issues.

It also adds many tests, including broken ones.

[mtfishman]

What parts of this PR rely on #1363? It would be simpler to review and merge this if it was separate from #1363.

[ogauthe]

replaced by #1531