LinearIndexing for a zero-dimensional view of an OffsetVector doesn't work as expected

[jishnub]

For a zero-dimensional array a we usually expect a[] == a[1]. However this does not work with zero-D views of OffsetVectors

julia> a = OffsetArray(1:3, 0:2);

julia> b = @view a[0]
0-dimensional view(OffsetArray(::UnitRange{Int64}, 0:2), 0) with eltype Int64:
1

julia> b[]
1

julia> b[1]
2

This happens because for FastContiguousSubArrays, the indexing produces b[1] -> a[1]. Perhaps the offset calculation for the view needs to be different for OffsetVectors?

It appears adding the method

Base.compute_offset1(parent::OffsetVector, stride1::Integer, I::Tuple{Int}) = I[1] - stride1

fixes the issue in this particular case, however I'm not sure if this holds in general. I wonder if this is the correct resolution? In that case I can consider a PR.

[johnnychen94]

Very nice catch! I think this is an edge case described in #100.

#101 fixes the issue described in #100 but doesn't cover this "edge" case.