Compatibility with SparseArrays

[kalmarek]

Hey, thanks for all the recent work on IntervalArithmetic!

It seems that common functionality for SparseArrays are still broken on 0.22.7 (despite #604):

julia> using IntervalArithmetic

julia> using SparseArrays

julia> x = [1.0 0; 0 0]
2×2 Matrix{Float64}:
 1.0  0.0
 0.0  0.0

julia> spx = sparse(x)
2×2 SparseMatrixCSC{Float64, Int64} with 1 stored entry:
 1.0   ⋅ 
  ⋅    ⋅ 

julia> spix = interval.(spx)
2×2 SparseMatrixCSC{Interval{Float64}, Int64} with 1 stored entry:
 [1.0, 1.0]_com        ⋅       
       ⋅               ⋅       

julia> spx .= zero(eltype(spx))
2×2 SparseMatrixCSC{Float64, Int64} with 1 stored entry:
 0.0   ⋅ 
  ⋅    ⋅ 

julia> spix .= zero(eltype(spix))
ERROR: ArgumentError: `==` is purposely not supported for intervals. See instead `isequal_interval`
Stacktrace:
 [1] ==(::Interval{Float64}, ::Interval{Float64})
   @ IntervalArithmetic ~/.julia/packages/IntervalArithmetic/7FL3g/src/intervals/real_interface.jl:83
 [2] fill!(A::SparseMatrixCSC{Interval{Float64}, Int64}, x::Interval{Float64})
   @ SparseArrays ~/.julia/juliaup/julia-1.10.0+0.x64.linux.gnu/share/julia/stdlib/v1.10/SparseArrays/src/sparsevector.jl:2303
 [3] copyto!
   @ Base.Broadcast ./broadcast.jl:964 [inlined]
 [4] materialize!
   @ Base.Broadcast ./broadcast.jl:914 [inlined]
 [5] materialize!(dest::SparseMatrixCSC{Interval{Float64}, Int64}, bc::Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{0}, Nothing, typeof(identity), Tuple{Interval{Float64}}})
   @ Base.Broadcast ./broadcast.jl:911
 [6] top-level scope
   @ REPL[24]:1

which can be traced to this line:
https://github.com/JuliaSparse/SparseArrays.jl/blob/4cc31f23619cfe4c2bcded16e71b7ec5cef974f9/src/sparsevector.jl#L2381

Looking at the comments of the file this is intentional to support -0.0 which iszero but is not equal to 0.0.

[Kolaru]

That's a case where [-1, 1] == 0 should return false, as per the standard. I never thought I would find one in my lifetime :'-)

On the downside, it means we can not fix it by changing == or iszero (it would break too many other codes). This will have to be a package extension.

[OlivierHnt]

On the other hand, on the 0.22 branch we have iszero(interval(-1, 1)) == false and interval(-1, 1) != 0.
For consistency we ought to allow interval(-1, 1) != interval(0).