OffsetArray broadcast of sparse vector tries to construct dense array

[Keno]

julia> using SparseArrays

julia> using OffsetArrays: Origin

julia> svec = SparseVector{Int, Int}(div(typemax(Int),2), Int[], Int[])
4611686018427387903-element SparseVector{Int64, Int64} with 0 stored entries

julia> svec .- svec
4611686018427387903-element SparseVector{Int64, Int64} with 0 stored entries

julia> Origin(0)(svec) .- Origin(0)(svec)
ERROR: invalid Array size

[jishnub]

Seems related to

OffsetArrays.jl/src/OffsetArrays.jl

Lines 283 to 287 in a094f04

	# TODO: Ideally we would delegate to the parent's broadcasting implementation, but that
	# is currently broken in sufficiently many implementation, namely RecursiveArrayTools, DistributedArrays
	# and StaticArrays, that it will take concentrated effort to get this working across the ecosystem.
	# The goal would be to have `OffsetArray(CuArray) .+ 1 == OffsetArray{CuArray}`.
	# Base.Broadcast.BroadcastStyle(::Type{<:OffsetArray{<:Any, <:Any, AA}}) where AA = Base.Broadcast.BroadcastStyle(AA)