Simplify rendering pipeline, reduce gpu memory usage, remove duplicated data

[PPakalns]

No description provided.

[grovesNL]

Thank you! I'd like to do some performance testing (especially WebGL) before merging this though. As far as I remember instancing small vertices used to be slower on older GPUs which is why I didn't do it this way from the start.

[PPakalns]

Didn't know that, learned something new :)

[hecrj]

I tested these changes in a relatively big app and I see a noticeable speedup on my M1 (from ~2ms frame time to ~1.7ms). I guess it's expected since the prepare loop does quite less. GPU usage seems unaffected.

[grovesNL]

I just rebased this onto main. Let's go ahead with this for the CPU time improvements.

It's possible it could regress GPU time some specific older hardware that doesn't handle small instances well, but most modern hardware should be ok with this based on what I can tell. e.g., it's confirmed that AMD has optimizations for small instances now instead of doing the slow thing that used to happen.

If we see a noticeable performance problem with any hardware then we can always special case certain vendors and expand the vertices out in those cases.