Optimize svo implementation

[jotabulacios]

This PR optimizes the protocol implementation.

The following optimizations were implemented:

Parallelized the main accumulator loop to distribute the heavy workload across all available CPU cores.

Eliminated heap allocations in hot loops by refactoring functions (compute_p_beta, etc.) to use mutable, stack-allocated buffers instead of returning new Vecs on each call. This significantly reduces pressure on the memory allocator and removes contention between threads.

Introduced a polynomial transposition pre-computation step (transpose_poly_for_svo) to ensure sequential memory access. This dramatically improves CPU cache utilization by converting a slow, strided memory access pattern into cache-friendly contiguous reads.

Results

Initial run

Size	Classic (avg median)	SVO (avg median)	SVO / Classic Ratio
16	2.87 ms	3.21 ms	1.12× (SVO slower)
18	9.87 ms	11.26 ms	1.14× (SVO slower)
20	33.51 ms	42.55 ms	1.27× (SVO slower)

After optimizations

Size	Classic (median)	SVO (median)	SVO / Classic Ratio
16	2.82 ms	2.68 ms	0.95× (SVO faster)
18	9.70 ms	9.42 ms	0.97× (SVO faster)
20	35.85 ms	33.27 ms	0.93× (SVO faster)