perf(sparse_strips): Reuse flattened_cubics allocation

[taj-p]

By reusing this allocation, we see 5 to 6.6% improvement to flattening performance.

[LaurenzV]

Is it not possible to pass the context all the way down and then access the buffer from the context directly instead of passing it separately? This way we might be able to avoid the mem take and putting it back later on.

[taj-p]

Is it not possible to pass the context all the way down and then access the buffer from the context directly instead of passing it separately? This way we might be able to avoid the mem take and putting it back later on.

Good idea! Thank you - implemented in 401e9fb