There are a bunch of interrelated changes in here that I couldn't really
separate out.  Mostly the relate to saving data on the AST nodes: we store
DerefInfo and closure info on the Name nodes, and store FunctionMetadata objects
on AST nodes that create scopes.  And those FunctionMetadata objects now also
store "ScopingResults", which is only the results from the ScopeInfo objects,
not all the extra info those keep around in order to do deferred analysis.

Once all this extra data is precomputed+stored, we can start freeing
the ScopingAnalysis objects and all the related ScopeInfos.

Future work:
- Name node shouldn't store scoing info; this should be embedded in choice of BST nodes
- Do pyc caching post-CFG
- Memory management.
- Continue to clean this all up.  I layered this change on top of the existing
  scoping and cfg systems, which reduced the amount of changes, but means that
  there is a potentially-unnecessary layer in the middle now.

Fix pointer reuse problems

while this removes some disabled features I'm not a big fan of having disabled stuff around
because it will break and makes development harder.

this is a duplicate - we will allready emit a call to recordType inside the bjit

solves a pointer reuse problem

Remove libpypa parser

After remove libpypa, the function getMagic only return one string. So
just use a char array constant instead a function.

MCJIT: free loaded objects

This fixes the llvm tier memory consumption until we migrate to the ORC JIT which has better memory management.

Misc memory size reductions

this does not hurt perf but still saves some memory

Threading fixes to get cffi-1.7 working

Instead of having it always point to the thread-local threadstate object,
it's a global that gets updated during context switches, like CPython does it.

There are still a number of differences that should be refactored out, but
I think this should bring us quite a bit closer to CPython's implementation.

It's time :(

This new version is causing some issues for us

switch from the SSE prefetch to __builtin_prefetch

string interning: make it slightly more space efficient

hidden classes: free classes of singleton type

this fixes the huge (~350MB) leak in:
for i in xrange(1000000):
    class C(object):
        pass

We used to store the string content twice.
This implementation makes use of DenseSet::find_as functionality where one can search using a different type.
I was not sure if the special DenseMapInfo I had to create causes any problems so I choose to create a new source file
for it so that it does not get picked up somewhere else.

free bjit code after LLVM codegen

we free now the code blocks after a recompile in the LLVM tier (except for OSR frames) because
it is likely that we will not use the code anymore.
- we have to make sure we are not currently executing any code we will delete that's why I added bjit_num_inside
- there were some cases where we forgot to deregister stuff
- when profiling we don't actually unmap the code in order to not brake profiling

EH: correctly deregister EH frames and cleanup the code