Change PyIterator::from_object` to return underlying TypeError

[davidhewitt]

Reviewing #1050 I realised that PyIterator::from_object() makes an unneccessary PyIter_Check() call as a successful result is already guaranteed to be an iterator.

I simplified it so that the function now returns the error from PyObject_GetIter() instead of a PyDowncastError. This helps the implementation of #1050 and I think also improves the experience here.

Now this function will return the well-known TypeError: 'Foo' object is not iterable which comes from PyObject_GetIter(): (https://github.com/python/cpython/blob/8182cc2e68a3c6ea5d5342fed3f1c76b0521fbc1/Objects/abstract.c#L2624)

[sebpuetz]

Just commenting on this since this was mentioned in my PR: #558 introduced this null check because trying to iterate some non-sense would cause segfaults. Is that behaviour fixed in some other way or would this lead to a regression here?

[davidhewitt]

Thanks for double-checking! Python::from_borrowed_ptr_or_err makes the null check inside it, so that case is still handled.

It actually happens exactly that the test I added to this PR tries to iterate an int and confirms that it results in a TypeError 😄