The organization is kind of hacky since we have to copy str_rsplit
into stringobject.c so it can be use stringlib, but we want to reference
it from str.cpp.

add support for relative imports, and bring our import api closer to cpython

copy _PyInt_Format implementation from cpython

add an explanatory comment about why we set level = -1 if we're missing `from __future__ import absolute_import`

Add the io/_io Module and make import gzip work

Treat unicode names as equivalent to their ascii-encoded str
version, and any non-ascii unicode name as a TypeError.

My hope is that this covers the common cases but doesn't cause
bugs.  You can see that we're doing this by putting unicode in
and getting a str out, but I hope that this is ok.

This adds quite a few C API functions
and I had to introduce atleast two temporary hacks:
- do not depend on the ABC module which uses weak references
- use a dummy PyErr_CheckSignals() function

in regards to how the metaclass gets picked.

'buffer' is currently just an empty class.

add weakref support for user defined subtypes (with inheritance) and 'type' instances

There are some times that the importing logic has to look up attributes
on objects, and we were using Box::getattr() (equivalent to
obj.__dict__[attr] ) instead of getattr() (equivalent to getattr()).
This was working fine until we started trying to run the custom import hook
in six.py, since that creates a subclass of ModuleType which defines
some of the important attributes as class-level attributes.

We apparently never tested compiling from_cpython files
with gcc, unless you manually passed USE_CLANG=0 which I
doubt anyone did.

Add a pyston_gcc target that just defers to cmake's better
ability to use different compilers, and run the test suite
through that rather than just pyston_prof.  This gives us
better coverage and also allows the Makefile to fully defer
`make check` to cmake.

Also, fix a couple bugs that gcc exposed.

cmake ignore gcc warnings in from_cpython

-added -Wno-unused-result and -Wno-strict-aliasing for gcc 4.8.2 compatibility

There are still a couple subtle differences in how we track and
store the various module attributes.  For example, we show something
different for 'print ModuleType("a", "b")'.

Also sneak in a fix for "import a.b.c as c"

Add should_error directive, for tests that should exit non-0

runtime/inline add header dependencies

-fixes #339