Trial using event flags for triggering shutdown routines

[reticulatedpines]

On ML shutdown we want to run various pieces of cleanup code, scattered across the code base. The core of this is in ml_shutdown(), which sets a global variable.

This is not very reliable or convenient, since code must poll for change of that variable. And if you're waiting on a sem, for example, you won't wake in order to check the global. This also relates to TASK_LOOP, which hides a check against ml_shutdown_requested. This is itself ugly; the intent is hidden. And most uses of TASK_LOOP do no cleanup code when the loop exits, so it's pointless.

This is probably why miscellaneous functions are called directly by ml_shutdown().

DryOS has a mechanism for low-latency, simple message broadcast; event flags. It should be cleaner, more reliable, and more efficient, to replace most checks against ml_shutdown_requested with waits for changes to a shutdown event flag. ml_shutdown() sets a flag, any cleanup code immediately triggers.