Centralized texture management does not prune unused textures when a platform allocation fails

The centralized texture management mechanism provided by the fix for RT-25323 provides proactive texture cleanup when we are about to go over a predetermined internal resource limit, but on a multi-user system we can also run out of the VRAM resource when other applications use too much VRAM and we can independently get an allocation failure from the system even though we are within our self-imposed internal limits. We need to invoke the necessary cleanup strategies when the system detects an allocation failure.