msm: kgsl: Correctly handle gmu fault interrupts

Send NMI to gmu so that we don't lose critical gmu state.
We do not want to trigger dispatcher because we don't know
whether the gpu was active or not. The best course of action
is to wait for the next kgsl -> GMU interaction to timeout
thus triggering GMU snapshot and appropriate recovery steps
based on whether gpu was active or not.

Change-Id: I17b4245f4e0113bfc902d7dae46bb24d0bc2b65d
Signed-off-by: Harshdeep Dhatt <hdhatt@codeaurora.org>