tty: n_smux: Add RX workqueue to reduce locking

To prevent multiple RX operations from overlapping, locking is currently
used during the entire RX operation which also includes calling the
client get_rx_buffer callback in atomic context.

To reduce locking, the RX processing has been moved to a single-threaded
workqueue which ensures serialization of the RX operations.  This
minimizes locking requirements and eliminates locking around
get_rx_buffer allowing the client to use GFP_KERNEL instead of
GFP_ATOMIC for memory allocation.

Change-Id: Ib87523191aa77f899fadb7667def58d94579c547
Signed-off-by: Eric Holmberg <eholmber@codeaurora.org>
Signed-off-by: David Brown <davidb@codeaurora.org>