Need some design for EGL context management

[RAOF]

We have a bunch of threads, a bunch of EGL contexts, and a bunch of methods which either assume a current GL context or setup a context on entry and tear it down on exit.

This is both doing unnecessary work (unbinding an EGL context implies a GL flush) and is a source of potential pitfalls - an EGL context can only be current on a single thread (so if we happen to need to access a context from multiple threads, it will fail), and there can only be one context current on a thread (so when we do the make_current on entry and release_current on exit dance we may have broken the caller).