improve renderer performance

- [ ] glium renderer currently builds the vertex list in a geometry shader which is probably not optimal.
- [ ] wgpu_renderer currently sends a ton of duplicate data to the gpu
- [x] for both renderers, we should be able to build just one vertex / index buffer per frame, then use sub-slices for individual draw calls.
- [ ] It might be possible to use multi-draw indirect to issue all draw commands simultaneously.

**Longer term ideas:**
- [ ] if we can unify renderers to use exactly the same raw vertex data, we can probably separate out the vertex data and build it as we go, essentially merging the widget build and render into a single pass. then, we would just need to build the buffers and execute draw calls.
- [ ] clip (and for that matter, all vertex) data sent to the GPU could be compressed from f32 into i16 probably.