Ensure only one update timeout is scheduled at a time and call gl.flush() to encourage eager execution

[mrxz]

While investigating the way CPU/GPU work is scheduled in Spark I noticed that the GPU work arising from SparkRenderer.updateInternal was not immediately executed, but instead waited for the next rAF or flush. See the following screenshot:

This PR introduces a gl.flush() to encourage eager execution. See WebGL_best_practices#flush_when_expecting_results. The browser can now immediately start the GPU work. Additionally the logic was reworked to:

• Ensure there's only a single timeout scheduled. Multiple timeouts would be scheduled in case >1 rAF took place before it got a chance to run or during WebXR sessions (once for each eye).
• Use a single THREE.Matrix4 instance to avoid cloning matrixWorld each time (avoids a memory allocation).

With these changes the timeline looks as follows:

Edit: for clarity the "GPU idle" and "GPU work" refer to the GPU thread. The key is that no GPU work is scheduled in the before case, whereas it is with this PR. Whether or not the GPU is actually idle at that point obviously depends on other factors. But in case it is, it can immediately start on it.

[dmarcos]

Great find! Thanks!