Before was using some bad iteration strategies for
fixed-point computation: LIFO for definedness analysis
(very bad), and FIFO for type analysis (bad).

Switch both of them to use a priority queue that picks the
block with the lowest index; this is an order of magnitude
better on the new gcj test but still is fairly slow.

Added some string functions and fixed a couple bugs.

Also, give up on maintaining 100% error message compatibility
with CPython.  There are some places where it's kind of silly
to do this since the messages seem based on internal implementation
details; namely, whether to output at the global scope:
NameError: global name 'x' is not defined
vs
NameError: name 'x' is not defined
This depends on whether any function has been defined that has
a "global x" declaration.

Instead, make the tester a little bit smarter to know that certain
classes of error messages should be treated as equivalent.

Not quite as small as microbenchmarks, which primarily
test just a single feature, but not full benchmarks either.

Will add them back later

Instead of releasing into the public domain, release
under the license of the respective upstream project.

Updated exception handling in:
	src/runtime/types.cpp
	src/runtime/builtin_modules/builtins.cpp
	src/runtime/builtin_modules/math.cpp

Adds a non-optimal implementation of int.__divmod__ + tests

Change the logic to determine path to parse_ast.py, otherwise it breaks usage of pyston binary.

There's still vestiges of the libunwind experiment; I'm leaving those in even
though they're dead since I think we should move soon back to that approach.

overlooked this one in the previous commit

Changed the indentation of pretty much the entire codebase.

It did some things that I don't like that seem not configurable,
but overall it seems like an improvement, and nice to have a
canonical format going forward.

Will commit the result in the next changeset.

Also, had to update to a newer version of LLVM to get some new clang-format options.
We don't necessarily have to use the same version of clang-format as the version of
llvm that we build it against, but if we can it keeps the build system simpler.

Release llvm_patches and libunwind_patches into the public domain so they can be incorporated into the upstream projects and their users

Uses module name when generating IR, for improved readability.

As I touched the module repr, I added some tests to it. Turned out that cpython normalizes the path of a module and uses the pyc file in the repr string. That part is ignored in this test.

Performance optimizations

reduces runtime of fannkuch.py by about 0.8 secs

Reduces fannkuch.py runtime on my machine from 25.5 secs to 16.5 secs.

Add support to list.count