arm/arm64: Remove spinlocks from on_cpu_async

Remove spinlocks from on_cpu_async() by pulling some of their
use into a new function and also by narrowing the locking to a
single on_cpu_info structure by introducing yet another cpumask.

Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
Acked-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>