Messages queues must be guarded on both ends or else it's a race between detecting...

Messages queues must be guarded on both ends or else it's a race between detecting a message present and missing a wakeup on thread about to wait. Keeping IRQs from interacting with the scheduler would be preferable but this should do at the moment. Add more detailed panic info regarding blocking violations so we know who. Make panicf function well enough on Gigabeat and PortalPlayer targets. Move the core sleep instructions into a CPU-specific inline to keep thing organized.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@13374 a1c6a512-1295-4272-9138-f99709370657