v1.6.0

• Renderer now lives on its own thread.
  This is a major architectural change that deserves far more of an
  explanation than I'm going to provide here.

  I've been wanting to split the renderer onto it's own thread for quite some
  time, mostly as an academic exercise.

  It turned out to create more maintenance burden than I would have liked,
  although it does make a few operations easier/cleaner, such as freeing GPU
  resources from worker threads.

  The API is a message-queue based system for which most of the boilerplate is
  generated by poof. There's an asyncify_render_function poof routine that
  takes a function prototype and spits out an Async version of the function
  which pushes a render job onto a work queue. At this point it's up to the
  programmer to remember to go actually call the function in the renderer
  (which is made easier by being reminded of the new, non-enumerated case by the compiler).

• Debug system once again welded directly into the program instead of compiled to a DLL
  I got tired of maintaining globals in so many places and decided the cost of
  having the debug system split out into a DLL was not worth the flexibility
  of being able to load it in other programs. If I want to use the profiler
  in other programs, I'll just compile it in.

• Special, compressed vertex data type for world_chunk
  I wanted to see how much performance I could squeeze out of compressing
  position and normal data for chunks to 8 bit int from 32 bit float. Turns
  out not a massive amount (deferred renderer is mostly texture-bound), but
  it did shave off a bit.

• Many other misc fixes
  I didn't document them along the way, and wrote this months after the work was
  done, so .. who knows.