ARC: [SMP] optimize IPI send and receive

* Don't send an IPI if receiver already has a pending IPI.
  Atomically piggyback the new msg with pending msg.

* IPI receiver looping on xchg() not required

References: https://lkml.org/lkml/2013/11/25/232


Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>