The defer backing memory pressure check gives immediate backing to
allocations made while local memory usage is low. Mid thread preemption
surfaces are created for every compute context at device initialization
and hardware writes them only when it preempts a thread, so each process
took local memory for them on every device, also on devices it never
uses. When defer backing is enabled, defer backing of preemption
surfaces without applying the memory pressure threshold.

Related-To: GSD-13830

Signed-off-by: Neil R. Spruit <neil.r.spruit@intel.com>
Source: 9346a08