event queue optimizations

[tehn]

notes from @catfact extracted from readme

• there is really too much allocation and mutexing going on in the event queue.
• allocation : posting an event invokes calloc to get an event_data union. since it is a union, its a good candidate for object-pool model.
  - threads : each read and write to the event queue is protected by a mutex. it is probably not a big deal, but would be better to have at least lock-free reads. (not sure how possible this really is)

[catfact]

coming back to this after doing some related work with lockfree SPSC queues, i don't think they are a good fit for this situation - they will tend to increase contention. since we don't have hard realtime deadlines, the mutexes are fine and should have insignificant overhead when contention doesn't happen.

the problem of course is when contention does happen - this would be when a script is doing some blocking operation during an event handler. and that just shouldn't be allowed to happen (see issue #461)

---

however, the allocation when posting events really is bad! events should use an object pool.