Switch to CPython's descrobject.c

The behavior changed in CPython 2.7.4, and Travis-CI runs 2.7.3.

until I realize that it's because we were passing more tests than we expected.

Create a simple Dockerfile

This is for adding a guard on a non-immortal object, since we need
to be safe against that object getting deallocated and a different object
is allocated in its spot.

We had support for this already, but it leaked memory.  The biggest was
that we never freed our runtimeICs, so if those ended up getting any GC
references in them, then we would leak memory.  So I started freeing those,
but then that exposed the issue that the ICInvalidators expect that their
dependent ICs never get freed.  So I added back a mapping from ICSlotInfo->
ICInvalidators that reference them.

via their descrobject.c

Not using it in this commit, just wanted to
get the unmodified version in so it's easier to see the changes.

We were doing a "call bumpUse() early" optimization to free up registers
when we can, but as a side-effect it looked to the refcounter like the
reference was done being used.

The problem is that we emit an llvm "unreachable" instruction, and then
continue to emit other code, which fails the verifier.

endBlock(DEAD) is supposed to be the right way to handle that, but there
is some more work that would need to be done there to get that working
properly.

So just do the easy thing for now -- create a new BB so that it's ok to
emit more code.

PyObject_New: register the type if the type is not yet registered

that is just based off of pulling our latest release

str: use tp_as_sequence instead of tp_as_number

string is special in that it is a c++ type which has tp_as_number and tp_as_sequence.
This causes problems because when we fixup the slot dispatcher we will set the tp_as_number fields but not the
tp_as_sequence because setting both can cause problems.
Some extensions (e.g. numpy) require that we use the sq_* functions instead of nb_*.
Therefore clear the tp_as_number fields (except nb_remainder which cpython has set too because it is not part of
tp_as_sequence).

BoxIteratorGeneric call PyIter_Next() lazily

update virtualenv to the one which always sets zip_safe=False

We were fetching to eagerly which caused issues if the iterator got used for other stuff too

Update test results

Merge conflict between my pr that added the full test suite and Sun's that
fixed one of the tests.

Some work on the numpy test suite

[WIP] Some minor numpy fixing.

binop: use PyNumber_* for user defined classes

microptimization: mark _PyGC_generation0 as having a const address

I think it's most important that we do this for the first arg,
since CPython won't end up increffing it either.

Looked at all places it comes from and this looks safe
(it's alive at least as long as the other arguments).

We are now crashing on cpython_oldstyle_getattr_crash (a CPython crasher)
but that's not really the fault of this change.

Merge refcounting into master

this is an advantage for _PyObject_GC_TRACK
which before compiled to something like this:
        cmpq    $-2, -16(%rdi)
        jne     .LBB0_2
        movq    $-3, -16(%rdi)
        movq    _PyGC_generation0(%rip), %rax
        movq    %rax, -32(%rdi)
        movq    _PyGC_generation0(%rip), %rax
        movq    8(%rax), %rax
        movq    %rax, -24(%rdi)
        leaq    -32(%rdi), %rcx
        movq    %rcx, (%rax)
        movq    _PyGC_generation0(%rip), %rax
        movq    %rcx, 8(%rax)

and now compiles to
        cmpq    $-2, -16(%rdi)
        jne     .LBB0_2
        movq    $-3, -16(%rdi)
        movq    _PyGC_generation0(%rip), %rax
        movq    %rax, -32(%rdi)
        movq    8(%rax), %rcx
        movq    %rcx, -24(%rdi)
        leaq    -32(%rdi), %rdx
        movq    %rdx, (%rcx)
        movq    %rdx, 8(%rax)

notice that it had to load _PyGC_generation0 three times because the compiler had to assume that because of aliasing it got motified.
(and this loads actually showed up in profiling as somewhat expensive)

tuple: implement tp_as_sequence->sq_contains