convert(Vector{Int}, o) is much slower than copy(PyVector{Int}(o))

As discussed [on julia-users](https://groups.google.com/d/msg/julia-users/1M8wDT0e5h4/qZ2OPaM9Hi4J), the two pathways to converting a Python list to a `Vector{Int}` seem mysteriously different in performance. For example:

``` julia
iv = PyVector(rand(Int16, 10^6))
@time copy!(Array(Int, length(iv)), iv);
@time convert(Vector{Int}, iv.o);
```

yields 3-5× slower performance for the second version on my machine.

Start by reproducing the issue's two benchmarks with a PyVector created from rand(Int16, 10^6): copy! into Array(Int, length(iv)) and convert(Vector{Int}, iv.o). Trace the two conversion paths to identify why their performance differs; done means the slower conversion no longer has the reported 3–5× gap, with the benchmark confirming the result.

Written by the indexing model from the issue text.