Fibril/async implementation overhaul.
This commit marks the move towards treating the fibril library as a mere
implementation of a generic threading interface. Understood as a layer that
wraps the kernel threads, we not only have to wrap threading itself, but also
every syscall that blocks the kernel thread (by blocking, we mean thread not
doing useful work until an external event happens — e.g. locking a kernel
mutex or thread sleep is understood as blocking, but an as_area_create() is not,
despite potentially taking a long time to complete).
Consequently, we implement fibril_ipc_wait() as a fibril-native wrapper for
kernel's ipc_wait(), and also implement timer functionality like timeouts
as part of the fibril library. This removes the interdependency between fibril
implementation and the async framework — in theory, the fibril API could be
reimplemented as a simple 1:1 shim, and the async framework would continue
working normally (note that the current implementation of loader complicates
this).
To better isolate the fibril internals from the implementation of high-level
synchronization, a fibril_event_t is added. This object conceptually acts
like a single slot wait queue. All other synchronization is implemented in
terms of this primitive.
