Performance: direct to headset blit

[avaer]

Right now Exokit has a multisampled texture framebuffer in the pipeline, which makes the render stack look like:

user render -> ms framebuffer -> blit to headset framebuffer

However, the blit phase is not necessary in many cases (such as a single page being rendered), because we can draw to the headset framebuffer directly, saving a blit/memory copy in the pipeline. This matters more on mobile, where memory bandwidth and gpu clocks are less and drain more battery.

The proposed architecture for the common case would be:

user render -> headset framebuffer

A code entry point:

exokit/src/index.js

Line 1076 in 3aa22c3

if (mlPresentState.layers.length > 0) { // TODO: composition can be directly to the output texture array