Investigate speeding up keyframe insertion

[HENDRIX-ZT2]

Currently, fcurves are updated for each key that is inserted. It would probably be faster to first precompute the final keys as a list and then insert them all in one go.
This could be a significant bottleneck for animation import.

https://blender.stackexchange.com/questions/92287/editing-fcurve-keyframe-points-in-fast-mode

Ideally, rewrite the key transforms to vectorize them with numpy so as to use less overhead. Especially interesting if the new codegen gives the data as arrays already.