and friends.  Should be pretty much the last step for letting
calls go all the way through our runtime using CAPI exceptions.

add __doc__ attribute to capifunc

Refactor location and names of gc visit functions.

This serves two purposes.
1) Improve naming consistency (follow the format TYPENAME::gcHandler)
2) In the source file, the visit function declaration is closer to the
   fields of the objects. If a field is added, hopefully it will be
   easier to remember to update the GC visit function.

Add "doctest" support

My guess is that there's a build dependency issue, where
we try building files that include libunwind.h before we
run the rules that generate it.  Not sure why we weren't running
into that before.  But the normal c++ files in src/ correctly
depend on libunwind, but the ones in src/runtime/inline don't
seem to.

For good measure, don't include libunwind.h in our header files,
and just use forward-declaring.

ie what you create either on the repl (where we were close) or
with `exec compile(foo, bar, "single")` (where we were not so close).

With these changes, doctest seems to mostly be working.

This was stemming from our use of the HEAPTYPE flag for our builtins,
and cpython inferring "is builtin" from that.

Also, fix a couple other minor things that I ran into.

object.__str__ use capi slot in order to increase compatibility with cpython

With this change we print True for:
class C(object):
    pass
print C.__str__ is object.__str__

Cheetah depends on this behaviour (see https://bitbucket.org/pypy/compatibility/wiki/cheetah)

Move our integration tests to a separate repository

By default it just does `git submodule init`

Enable future builtins module and add test file

Hash improvement for complex, str and long.

section ordering hack: use the PGO-determined ordering

Rather than do anything smart, just copy the ordering that the
pgo build used.  To keep the file reasonable, only emit
entries for functions that appear in the perf results (cuts from
>1M entries to ~2k).

I tried getting the hhvm "hfsort" to work, but ran into some
fixable-but-annoying issues getting it integrated (ex need to bring
in some other libraries).  Let's just try this for now; the goal
isn't to get the absolute best performance anyway, since for that
we'll use PGO and ignore the section ordering file.

getiter: use tp_iter for non python types

Some improvements to make test unary could pass

Recursive printing

Enable `test slice` and add missing function.