Flesh out user locked/unlocked lifecycle.

When a user is first started, we assume that they're "locked" meaning
that credential-encrypted data is unavailable.  Once credentials have
been supplied, we can transition the user to a fully running state.

To facilitate this lifecycle, UserState now has two separate
RUNNING_LOCKED and RUNNING states.  To ensure consistent events are
sent on all devices, we always step through RUNNING_LOCKED before
arriving at RUNNING.  This consistency means that apps processing
data based on the new ACTION_LOCKED_BOOT_COMPLETED broadcast and
system services using the new onUnlockUser() event will be less
bug-prone over time.

If the user storage is unlocked (which is the case on the majority
of legacy devices), we immediately transition from the RUNNING_LOCKED
into the RUNNING state.

Add logging for all state transitions.

When we "recover" a user in the process of being shut down, return
to the last known state.

Bug: 25943941
Change-Id: I5fec980f10b0d0fb2c272a662d193dc15136f9b9