Array/vector comprehension and AbstractTrees Leaves iterator

Why the array comprehension behaves differently with the AbstractTrees.Leaves iterator?
In the following replacing Leaves(t) with, say, range(1,3,3), I do get the 2D array.

julia> using AbstractTrees

julia> t = [[1, 2], 3]
2-element Vector{Any}:
  [1, 2]
 3

julia> [(i,j) for i in Leaves(t), j in Leaves(t) ]
9-element Vector{Tuple{Int64, Int64}}:
 (1, 1)
 (2, 1)
 (3, 1)
 (1, 2)
 (2, 2)
 (3, 2)
 (1, 3)
 (2, 3)
 (3, 3)

julia> [(i,j) for i in collect(Leaves(t)), j in collect(Leaves(t)) ]
3×3 Matrix{Tuple{Int64, Int64}}:
 (1, 1)  (1, 2)  (1, 3)
 (2, 1)  (2, 2)  (2, 3)
 (3, 1)  (3, 2)  (3, 3)

Start by reproducing the shown Julia REPL examples with AbstractTrees.Leaves(t), collect(Leaves(t)), and range(1,3,3). Compare the iterator behavior with Julia's array-comprehension rules; done means establishing whether Leaves is incorrect or the behavior needs documentation, with a focused regression test or explanation.