Modify condition in MultiDimClosure function

[Zinoex]

For context, I am trying to load a Array{Vector{Int64}, 4} using JSON.jl, which relies on StructUtils.jl. The inner vector is jagged, i.e. not the same length for every element, so Array{Int64, 5} is not an option. Now, without this stop condition, the multidimensional array continues until f.cur_dim[] == 0 and fails to set the dimension, unsurprisingly. With this condition it works flawlessly.

Full disclaimer, I don't know if this will have a major impact elsewhere, so please review this PR carefully. But it would be nice if loading vectors in multi-dim arrays worked.

[quinnj]

Thanks for the fix! The change looks correct - without this condition, loading jagged arrays like Array{Vector{Int64}, 4} would fail when the innermost arraylike element is reached.

Could you add a test case that covers this scenario? Something like:

# Test jagged nested arrays (Array of Vectors)
jagged = [[1, 2], [3, 4, 5], [6]]
result = StructUtils.make(Vector{Vector{Int}}, jagged)
@test result == jagged

# Or a multi-dimensional case
md_jagged = Array{Vector{Int}}(undef, 2, 2)
md_jagged[1,1] = [1, 2]
md_jagged[1,2] = [3]
md_jagged[2,1] = [4, 5, 6]
md_jagged[2,2] = [7, 8]
# test round-trip through make

Once a test is added, this looks good to merge!

[Zinoex]

@quinnj Adding the tests revealed that discover_dims also required modifications. While not part of the API, downstream packages may have chosen to add a discover_dims for their specific style in which case this could be considered breaking changes. A cursory Github search tells me that only StructUtils (and forks) provide discover_dims methods but I cannot be certain.