rtt memory allocation & lazy loading

[wouterlucas]

Currently if you use RTT it loads everything regardless of its visibility state, meaning all RTT nodes are always loaded and it is counted double on the memory manager.

Ideally we want:

• RTT nodes to only load when they are inbound of their respective view port
• Memory counted should only account for the final texture generated, not the intermediate buffer used to generate the RTT node

Nodes loaded without RTT:

Same test with RTT:

From first glance at the Chrome task viewer it doesn't seem to increase the memory on the GPU, and that makes sense. Likely just double counted on the memory manager side. But TBD investigation on a device.