reuse DynamicVectors and recurse in parallel

[mikowals]

After seeing your project on Discord I ended up running the code through a profiler. Most of the time was being spent allocating the DynamicVectors. So I made a couple changes based off that.

• reuse DynamicVectors. Now they are only created in the initial values and when half_vector is called. To reuse appropriately an offset and size need to be passed around with the vector. When lifetimes are done and vector slices can be worked with the code can be simplified.
• fuse the two calls to half_vector. Calling a single for loop that reads the vector in order is more efficient than separate loops that read the vector with stride 2.

I also added parallelization. This is self explanatory except maybe for the thread limit. Because the function recurses I found it needed a limit to stop from crashing and also it could be tuned. I am on M1 Mac with 10 cores and 8 high performance cores. 16 or 32 seemed to give the best results.

The benchmark output time goes from 0.065 -> 0.0054 seconds based on these changes.

[duckki]

Thanks. This is great!