With some changes
- renamed AST_Exec::expr -> body (to match CPython)
- removed references to AST_SetComp

add two passes which remove unnecessary boxing

* the first one removes boxInt, boxFloat and boxBool calls where the argument is coming from a corresponding unbox call
* the second pass removes duplicate boxing calls inside the same BB

together they improve the performance by about 10%

If we had to guard on the type of an object but the variable
ended up being undefined, we treated that as a guard failure
(not sure why).

Fixing that exposed another issue: if we guard that an object
is an int, we will try to unbox it if necessary.  If the variable
wasn't defined, then we will try to unbox some garbage memory.
Use the new handlePotentiallyUndefined to deal with that.

Before, for compiled functions we used dwarf line number information.

Now it's switched to using the new frame introspection support, and
we pass through the AST_stmt* as one of the arguments.  This gets us
the line number and also the definition of execution point.

and exc_info -> unw_info

(This makes more sense with the next change, but I wanted
to separate this large renaming from the functional change.)

I wonder if other people have run into this.

Previously things were hardcoded to only allow a single entry point to
the interpreter (since that is how the unwinder determines if the frame
is an interpreter frame).

Instead of having astInterpretFunction be the interpreter frame, make
execute() the frame.  Have to make it a static class function so we can
take its address.

Can still switch back to the default regalloc using the -b flag,
but it looks like the basic register allocator cuts the LLVM JIT
time significantly, and improves every one of our benchmarks.

With the recent package import support, we broke importing optparse.
This is because we now can import "encodings" (a package), which imports
codecs, which raises a SystemError when trying to import _codecs which
we don't support yet.

This was getting ignored before since optparse wrapped this process
(which happens via "import gettext") in a try-except and catches the
ImportError.

Add missing FloorDivInt test

Cmake fixes for #231 and #232

load package/__init__.py if it exists, and set its __path__

my old one was missing a std::move and did not accept lvalues.

Controlled by the '-b' command-line flag.  That used to
be for "benchmark" mode but it looks like benchmark mode
has been broken for a while, so remove that.

also, add a test that includes importing both a module with __init__.py and a submodule.

initial cmake support

ie due to reoptimizing the function and then calling into the compiled code.
We could handle this specific case by not treating these as "real" frames
that need to be introspected, but we have these kinds of situations in other
places (reopt from compiled code, osr from either kind) so I'd rather
find a more general improvement.

Before it was recompiling the OSR-entry function every
time it would take an osr exit.