WIP DO NOT MERGE: Partial transferable support

[mlogan]

See #58

Ok, here's where I've gotten so far:

• TypedArrays and their buffers are serialized/deserialized properly, as in Chrome webworkers. (i.e. the thing that comes out on the other end is an identical typed array).
• Transferables work in that new TypedArrays are constructed in the other thread from the same memory that was held in the previous thread. Arrays that are transferred become inaccessible in the thread they were sent from.

Still TODO:

• I haven't tested this extensively, or put much effort into input validation in the postMessage API.
• THIS CODE LEAKS BUFFERS: I've tried to use a Persistent handle with a weak callback to release the ArrayBuffer memory when the ArrayBuffer is GCed. Once you Externalize an ArrayBuffer as we do here, v8 will no longer free the backing memory at GC time. (See https://groups.google.com/forum/#!searchin/v8-users/ArrayBuffer/v8-users/CtOe7f6znZU/CDm-Rn0oJcMJ and https://groups.google.com/forum/#!searchin/v8-users/ArrayBuffer/v8-users/whHPAIhfMu8/D5H2-dP0HU4J for more info). However, I have no been able to get the callback to fire despite explicit calls to gc() with --expose_gc. I'm probably doing something wrong. I've gotten WeakCallbacks working before, so I'm somewhat at a loss as to why they aren't working here.
• Additionally, even if we get the weak callbacks working, we need to relinquish our weak callback if the buffer is sent back to another thread. Otherwise double-deletes will result.
• I haven't made much effort to use Nan for v8 API usage. Much of this code has no chance of compiling on v0.10 or earlier anyway, as that predates the ArrayBuffer/TypedArray C++ API in v8.

I'm probably not going to spend much time trying to resolve these issues, as I'm doing this for an embedded application in which we've already made some changes to the v8 ArrayBuffer API that will make the GC difficulties go away anyway.

That's no help to anybody wanting to use this code with an existing node install of course, for which reason this PR should not be merged. I'm posting it here in case anyone wants to take a crack at fixing the GC issues.