SDF text rendering buffer refactor

[wouterlucas]

Currently the SDF text render uses a double buffer along side the main color buffer, this results into double buffer swapping and excessive draw calls.

This results into high performance issues on low-end devices, where SDF is expected to be much faster the performance drops off quickly as the current approach generates a lot of overhead. Creating "lag" when the text is being rendered.

Possible changes:

• Concat the SDF buffer to the main color buffer
• Render SDF text to a texture and copy it in
• Leverage instance drawing for text to reduce text overhead

cc @erikhaandrikman

[wouterlucas]

Working on POC that has 3 SDF buffers:

• Small SDF text buffer sdf-text-small which should be able to store up to 64 text regions
• Medium SDF text buffer sdf-text-medium which should be able to store up to 32 text regions
• Large SDF text buffer sdf-text-large which should able to store up to 16 text regions

Idea is we "cache" the SDF text in either of these 3 buffers, assuming its big enough for most UIs (to be validated) and render them out of the buffer by exact region when needed instead of creating individual framebuffers for each SDF text buffer node as in the current implementation.

This should reduce constant creation/deletion of framebuffers (which is slow on older devices). If the SDF buffers are full, we fall back to a locally created framebuffer (else we'd get no text).