mhi: core: Fix PM state machine sync issue

Certain corner cases have been identified in the MHI core power management
state machine whereby the driver state can go out of sync with the
device.

The cases are as follows: During in MHI resume operation the device
is not guaranteed to be fully recovered from LPM as the framework
requires. This causes RPM suspend to potentially be invoked
on a device which is still going through a resume state.

During RPM suspend action, if said action is cancelled a timing
window exists where previously cached commands are not played back
to the device. This causes a rotting command issue, when said commands
are never acknowledged causing a non recoverable timeout error on the
client channels.

During an RPM suspend operation, a timing window exists where
the device may be transitioning to M2 when the M3 is set. This causes
a device error which is non recoverable except by a full reset.

CRs-Fixed: 856202
Change-Id: I232365ba77b02b9aec87fef4cecc3d991243afe2
Signed-off-by: Andrei Danaila <adanaila@codeaurora.org>