New semantic analyzer: allow replacing submodule with a variable

[ilevkivskyi]

Although this looks like an arbitrary exception from general rules, this:

• is an actual pattern used by some libraries
• matches old analyzer in most cases
• matches runtime semantics

[JukkaL]

Even this is a bit ad hoc, it's useful to be able to support this use case. I have a few questions about test cases, otherwise looks good (assuming tests pass).

[JukkaL]

What about trying to import lib.config as a module? Will it work? What about import lib.config as c?

[ilevkivskyi]

Hm, OK, I will try adding these tests.

[ilevkivskyi]

OK so these two have different behavior in mypy:

• import lib.config actually imports lib, so when one writes lib.config later, the variable takes precedence.
• import lib.config as tmp finds module lib.config and binds it to tmp, so when one writes tmp later it refers to the module.

At runtime variable always wins. Fixing the second one is non-trivial, because we need to check whether there is a "shadowing" variable at every submodule level. Also this is kind of corner case to a corner case.

[JukkaL]

Would it be useful to an incremental mode test case as well?

[ilevkivskyi]

Yes, I will add an incremental test.