While a little more awkward to use, this gives us a bit more
flexibility. In particular it will let us make this variable resettable
after fork by adding a `RefCell`.

This will let us add interior mutability (to replace after fork)
cheaply; without needing some form of Mutex that needs to be locked
and unlocked on every access.

The initial implementation of fork won't support running with the memory
manager (`MemoryMapper`) enabled. Move it to its own test suite so that
we can continue to run the clone tests with it enabled.

Unlike when creating a new thread, when we create a new process the
child process finishes the shim-side syscall handling normally instead
of long-jumping back to the point of the original syscall. We handle
this on the shadow side by validating the native clone syscall result
and propating that back to finish the virtualized clone syscall.

…lStorage::unregister_current_thread

The fork libc function actually uses the clone syscall with
CLONE_CHILD_CLEARTID and CLONE_CHILD_SETTID, so those needed to be
implemented as well.

…cted_final_state