PR: Return types for CompiledObjects using docstrings.py

[bcolsen]

This pull request in conjunction with PR: #796 for numpydoc returns in docstrings solves issue #372 for numpy complied objects. This makes completions functions numpy.array to return an numpy.ndarray object resulting in returns like this:

>>> import jedi; jedi.Script('import numpy as np; np.array([1,2,3]).a').completions()
[<Completion: all>,
 <Completion: any>,
 <Completion: argmax>,
 <Completion: argmin>,
 <Completion: argpartition>,
 <Completion: argsort>,
 <Completion: astype>]

I have not yet written a test since I need to import numpy to do such. A standard python library with complied objects and sphinx documentation would be a better test.

[bcolsen]

This is necessary due to cyclical importing of CompliedObject in analysis.py which is import by docstrings.py through iteratable.py

[ccordoba12]

@bcolsen, could you update #796 and this one to work with the new 0.10.0?

[bcolsen]

@ccordoba12, I can look at getting the spyder jedi patch to work for both Jedi 0.9.0 and 0.10.0.

If needed, I could work on #796 but I'm not familiar with the googledocs stuff. I'm also going to wait on this PR until #796 goes through.

[ccordoba12]

@ccordoba12, I can look at getting the spyder jedi patch to work for both Jedi 0.9.0 and 0.10.0

Yes, that'd be great, thanks!

[bcolsen]

Closing this because I submitted PR: #868 that includes an updated version of this code.