[GarboMuffin]

Need to review this more but this slop seems promising

four optimizations:

• replace buffer subdata with fully new buffers. sounds unintuitive but actually seems to make it faster. the issue with subdata is if GPU is currently reading that array then this causes it to need to be copying it internally anyway
• don't unbind framebuffer when switching between pen lines <-> pen stamps. unbinding the framebuffer at all causes the GPU pipelines to be less efficient
• reduce some repeated setup for consecutive stamps
• manually inline very hot pen code to avoid function call

[GarboMuffin]

probably better to have a render region for "i am rendering to the pen" and then have the within-pen stuff negotiate whatever state they need. this probably breaks extensions rn