Handle nested class names in stubgen

[ryancahoon-zoox]

Correct stubgen's handling of nested types (e.g. a class defined inside another class).

Previously, stubgen generated import statements in the .pyi file which includes the outer type.

I fix this by testing that the module name in the generated import is a module by using importlib.util.find_spec. This is the approach recommended by the python docs: "If you need to find out if a module can be imported without actually doing the import, then you should use importlib.util.find_spec()."

If stubgen is not able to access the referenced module, the new functionality does have a graceful fall-back behavior: if none of the possible partitions of module_name and class_name result in a successful import, it will default to the previous behavior of assuming that just the last segment of the type name is the class name (and the rest is the module name).

This may add some performance overhead from needing to call importlib.util.find_spec on each referenced type, but I don't know of another way to handle this nested-type case, unless we rely only on some sort of naming convention to distinguish type names from module names.

A unit test for the nested-type case is included.

Closes #633

[wjakob]

Looks good to me, just a minor comment.