Wrong inference maybe due to inference cache

[hippo91]

This bug has been found during the setup of the numpy brains.
Fixing it could improve our inference system for numpy functions.

The directory described hereafter reflects a similar structure inside the numpy sources.

Steps to reproduce

1. Create a directory named bug_inf with the following structure:

$ tree bug_inf
bug_inf
├── core
│   ├── __init__.py
│   └── shape_base.py
├── __init__.py
└── lib
    ├── __init__.py
    └── shape_base.py

2. bug_inf/core/__init__.py content is:

$ cat bug_inf/core/__init__.py 
from .shape_base import *

3. bug_inf/core/shape_base.py content is:

$ cat bug_inf/core/shape_base.py 
def vstack():
    return 42

4. bug_inf/__init__.py content is:

$ cat bug_inf/__init__.py 
from .core import *
from .lib import *

5. bug_inf/lib/__init__.py content is:

$ cat bug_inf/lib/__init__.py 
from .shape_base import *

6. bug_inf/lib/shape_base.py content is:

$ cat bug_inf/lib/shape_base.py 
from bug_inf.core import vstack

7. Create a file bug_inf.py which content is:

from astroid import extract_node

SRC = """
from bug_inf import vstack
vstack()
"""

NODE = extract_node(SRC)
print(NODE.inferred())

8. Run python3 bug_inf.py

Current behavior

 $ python3 bug_inf.py 
[<Const.int l.2 at 0x7f08471338d0>, Uninferable]

Expected behavior

 $ python3 bug_inf.py 
[<Const.int l.2 at 0x7f08471338d0>]

Astroid version

python3 -c "from astroid import __pkginfo__; print(__pkginfo__.version)"
2.3.0

[hippo91]

Here is attached an archive with the files described in the issue.
bug_inf.tar.gz

[hippo91]

Fixed thanks to PR #884