Add more external projects to our travis-ci test suite

Use the padding bytes at the end of BoxedString

CPython has this optimization as well, and it should
save us an average of 7 bytes per string.

Some `__future__` import stuff

allocate runtime ics using mmap (with PROT_EXEC)

output stderr/stdout diffs for failed tests

free generator even if the don't exit

also, output both in KEEP_GOING mode if VERBOSE

Specify all package versions in virtualenv_test

We were doing this for the main packages, but we would get different
versions of their dependencies dependending on what the latest version
was.

Also, add some extra debugging output in the parser in case that
issue crops up again.

Try to make weakref4 less flaky

Slightly change generator StopIteration handling

Make multiple copies of all of the objects we expect to get
collected, and just assert than any of them get collected.

Unconditionally raise a StopIteration when trying to iterate
a finished generator, regardless of what originally stopped
it from iterating.

Generator: don't generate the traceback for the StopIteration exc when the generator is exiting back to a hasnext call.

Add the _elementtree and update libpypa

Remove redundant code and pave way for #254

Fix compile error when enabling memstats + call tp_richcompare even if rewrite_args == NULL

I think this makes sense and I having different behaviour depending on rewrite_args is strange.

unifying behaviour -- step 2

unifying behaviour -- step 3

Generator: don't generate the traceback for the StopIteration exc when the generator is exiting back to a hasnext call.
This is a small perf improvement for django-template.py: took  7.4ms for last iteration from 8.2ms

added some files that test_float needs

Optimize stattimers

Main change is separating the representation of the timer stack (StatTimer) from
the manipulation of it (ScopedStatTimer).  This makes it easier to keep the different
use cases (top level timers vs non-top-level) separate without having the individual
functions check whether they think they are a top-level timer or not.

I would have expected this to help a bunch but it seems to actually hurt

This roughly halves stattimer overhead, from +80% to +40%
(on a very small benchmark; more complex benchmarks probably
have lower overhead)