Fix sparse constructor when Tridiagonal/SymTridiagonal are empty

[mcognetta]

Constructing a sparse matrix from empty structured matrices works for some, but not all, structured matrix types. This PR fixes the cases of Tridiagonal and SymTridiagonal. Diagonal and Bidiagonal already work as expected.

---

Currently:

julia> sparse(Diagonal(zeros(0, 0)))
0×0 SparseMatrixCSC{Float64, Int64} with 0 stored entries

julia> sparse(Bidiagonal(zeros(0, 0), :U))
0×0 SparseMatrixCSC{Float64, Int64} with 0 stored entries

julia> sparse(Tridiagonal(zeros(0, 0)))
ERROR: ArgumentError: invalid Array dimensions
Stacktrace:
 [1] Array
   @ ./boot.jl:452 [inlined]
 [2] SparseMatrixCSC{Float64, Int64}(T::Tridiagonal{Float64, Vector{Float64}})
   @ SparseArrays ~/github/julia-dev/stdlib/SparseArrays/src/sparsematrix.jl:575
 [3] SparseMatrixCSC
   @ ~/github/julia-dev/usr/share/julia/stdlib/v1.8/SparseArrays/src/sparsematrix.jl:563 [inlined]
 [4] sparse(T::Tridiagonal{Float64, Vector{Float64}})
   @ SparseArrays ~/github/julia-dev/stdlib/SparseArrays/src/sparsematrix.jl:760
 [5] top-level scope
   @ REPL[10]:1

julia> sparse(SymTridiagonal(zeros(0, 0)))
ERROR: ArgumentError: invalid Array dimensions
Stacktrace:
 [1] Array
   @ ./boot.jl:452 [inlined]
 [2] SparseMatrixCSC{Float64, Int64}(T::SymTridiagonal{Float64, Vector{Float64}})
   @ SparseArrays ~/github/julia-dev/stdlib/SparseArrays/src/sparsematrix.jl:606
 [3] SparseMatrixCSC
   @ ~/github/julia-dev/usr/share/julia/stdlib/v1.8/SparseArrays/src/sparsematrix.jl:594 [inlined]
 [4] sparse(T::SymTridiagonal{Float64, Vector{Float64}})
   @ SparseArrays ~/github/julia-dev/stdlib/SparseArrays/src/sparsematrix.jl:758
 [5] top-level scope
   @ REPL[11]:1

This PR:

julia> sparse(Diagonal(zeros(0, 0)))
0×0 SparseMatrixCSC{Float64, Int64} with 0 stored entries

julia> sparse(Bidiagonal(zeros(0, 0), :U))
0×0 SparseMatrixCSC{Float64, Int64} with 0 stored entries

julia> sparse(Tridiagonal(zeros(0, 0)))
0×0 SparseMatrixCSC{Float64, Int64} with 0 stored entries

julia> sparse(SymTridiagonal(zeros(0, 0)))
0×0 SparseMatrixCSC{Float64, Int64} with 0 stored entries

[mcognetta]

Added a fix and tests for the 1x1 case.

[dkarrasch]

Let's construct the SparseMatrixCSC directly also in the m=1 case. While you're at it, could you please fix the new line 659 by

SparseMatrixCSC(m, m, Vector(Ti(1):Ti(m+1)), Vector(Ti(1):Ti(m)), Vector{Tv}(D.diag))

otherwise it doesn't respect the index type Ti.

[mcognetta]

Let's construct the SparseMatrixCSC directly also in the m=1 case. While you're at it, could you please fix the new line 659 by

SparseMatrixCSC(m, m, Vector(Ti(1):Ti(m+1)), Vector(Ti(1):Ti(m)), Vector{Tv}(D.diag))

otherwise it doesn't respect the index type Ti.

Fixed. Added a test for the SparseMatrixCSC{Tv, Ti}(::Diagonal) case as well to ensure Ti is respected.

[dkarrasch]

This should be backported to v1.6.