deprecate o[integer] so that we can transition to 0-based access

[stevengj]

Now that o.foo mirrors Python, we also want o[foo] to mirror Python. Unfortunately, PyCall currently makes o[integer] in Julia act like o[integer-1] in Python, to emulate 1-based access. This PR deprecates 1-based indexing of PyObject so that in PyCall 2.0 we can switch to 0-based indexing.

[carstenbauer]

Naively, to have to mentally switch between 0 and 1-based indexing depending on whether I'm accessing a Julia or Python object seems like a big inconvenience. Maybe I misunderstand though. Could you elaborate why you want to switch to 0-based indexing? Isn't consistent 1-based indexing just as good - and more user friendly?

[stevengj]

Because o[foo] will soon be equivalent to o[foo] in Python for all foo, not just integers. Imagine a Python dictionary, for example. Once we do that, it will be weird to modify foo only if foo is an integer.

Auto-conversion will still translate a Python list to a 1-based Julia array (or PyVector). But if you have a raw PyObject then I think it should mimic Python.