BN254 Block Multiplier (Rust Version)

[Quarky93]

This PR introduces a new crate called block-multiplier. The crate provides a primitive for efficient modular multiplication, designed to serve as a low-level building block for higher-level cryptographic operations such as the Number-Theoretic Transform (NTT) and the Skyscraper hash.

This is work done by the working group:
Yuval Domb --- @yuvalingo
Xander van der Goot --- @xrvdg
Koh Wei Jie --- @weijiekoh
Tony Wu --- @Quarky93

Motivation

The core idea behind block-multiplier is to maximize throughput of modular multiplications by exploiting the parallelism between the scalar integer and floating-point SIMD pipelines available on modern ARM CPUs. Specifically, this design targets:

3 modular multiplications per block:
    1 via the scalar integer pipeline
    2 via the floating-point SIMD pipeline
On optimized platforms (like the Cortex-A76 in the Raspberry Pi 5), the goal is for the latency of a block (3 multiplies) to match that of a single scalar multiplication.

Platform Notes

Raspberry Pi 5 (Cortex-A76): Block size of 3 seem to be optimal. Performance gain significant.
Apple Silicon (M1, M3): Slight tuning differences may still be required.

Algorithms
The techniques implemented draw from the following works:

Yuval Domb @yuvalingo --- [multi-precision modular reduction scheme with only n^2 + 1 digit multiplications](https://hackmd.io/@Ingonyama/Barret-Montgomery)
Niall Emmart @nemmart --- [Modular Exponentiation Using Double Precision Floating Point](https://ieeexplore.ieee.org/document/8464792)

These works were implemented by Xander van der Goot @xrvdg in the modmulzoo repo.

Further improvements planned, but will be a future PR.

[xrvdg]

I've changed the base branch from main into xr/rtz (#28) for easier review.

[recmo]

Interesting. Apparently if I delete the target branch (which was merge into main), it auto-closes any outstanding PRs. Furthermore, the 'reopen' button is disabled and I can not change the target branch to main. 🤔

[recmo]

ok, re-created the target branch.