If cpu_cluster_pm_enter() fails, we need to set MPU power domain back
to enabled to prevent the next WFI from potentially triggering an
undesired MPU power domain state change.

We already do this for omap_enter_idle_smp() but are missing it for
omap_enter_idle_coupled().

Fixes: 55be2f50 ("ARM: OMAP2+: Handle errors for cpu_pm")
Signed-off-by: Tony Lindgren <tony@atomide.com>