arm Neon vector backend

[Tarinn]

This adds a new vector backend that utilizes arm neon SIMD instructions to implement the parallel formulae, conforming to the avx2 implementation. Other files outside of of the new neon backend are updated to be able to use this backend when the required feature flag is set. The new backend is largely a copy of the avx2 backend with the instructions changed to arm neon, and other changes from the 'base' avx2 implementation mainly focus on the need to split a 256 bit vector into two 128 bit ones and differences in structure of input/output of arm neon instructions.

On a Raspberry Pi 4, the recorded speed-up of serial vs vector backend was around 20% for relevant benchmarks of variable-time multiplication.

[rubdos]

@Tarinn I think you should squash the last two commits into d822a2c, since that last one accidentally committed a bunch of large files that you removed later.