Improve flight mode tracing performance for large tracings

[daniel-wer]

This PR includes a couple of performance optimizations:

• Decrease UNDO stack size to 20 (from 100). For a ~80.000 node tracing the undo stack grew to ~1GB which may decrease tracing performance. If users want to restore earlier versions they can always use the "Version Restore View".
• Optimize adding a single edge to the edge_collection.
• Tweak the diffable_map size. Decreased from 10.000 to 1.000, which improved diffing/cloning time when adding a single edge (which is the most frequent case).
• Removed the existsCache from the diffable_map. Cloning this map was a major performance bottleneck. When setting we now no longer know whether the key already exists and need to search through all chunks, but that is much faster than the cloning. Also, much less garbage collection during the performance measurement (from up to 20% down to 5%). This change had the biggest positive performance impact :)

I tested this with an 80.000 node tracing. FPS during flight mode tracing increased from ~3-7FPS (and hefty lag) to ~17-22FPS, where smooth tracing is very possible :)
I also tested whether tree split performance was affected by the changes, which it was not (but it is and was rather slow).

Not sure whether this will also fix the crash that was reported, probably not :/

URL of deployed dev instance (used for testing):

• https://improveflightperformance.webknossos.xyz

Steps to test:

• Use the tracing from https://discuss.webknossos.org/t/task-very-slow-webgl-crash-for-very-large-tracing-in-flight-mode/1417 and trace in flight mode. Tracing with a reasonable movement speed should be possible (I used 200-300 nm/s).

Issues:

• fixes https://discuss.webknossos.org/t/task-very-slow-webgl-crash-for-very-large-tracing-in-flight-mode/1417

---

• Updated changelog
• Ready for review

[philippotto]

Excellent PR 👍 Thanks for investigating and improving this so thoroughly 🎉