Reductionless block multipliers

[xrvdg]

This PR removes the reduction step from the scalar and SIMD codegenerators. This gives the caller freedom to choose what kind of reduction strategy to employ.

Additionally, this adds code generators for a 2-round (log jump) Montgomery multiplier and squarer. Without the reducer, the output of the single-step squarer is too high to be used in the square operation of Skyscraper.

stacked on top of #108
closes #120

[codspeed-hq]

CodSpeed Walltime Performance Report

Merging #116 will degrade performances by 59.99%

_{Comparing xr/remove-reducer (50ecf99) with main (9a9656a)}

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

Summary

⚡ 8 improvements
❌ 6 regressions
✅ 16 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	ark_ff	6.3 µs	4 µs	+58.05%
❌	block_mul	4.1 µs	5.1 µs	-20.54%
❌	montgomery_interleaved_3	3.8 µs	4.6 µs	-18.12%
⚡	montgomery_interleaved_4	6.3 µs	3.9 µs	+60.57%
⚡	scalar_mul	6.1 µs	3.9 µs	+57.53%
⚡	simd_mul	7.1 µs	4.5 µs	+58.2%
❌	ark_ff	5.2 µs	6.6 µs	-20.96%
⚡	block_sqr	5.8 µs	3.7 µs	+56.92%
❌	scalar_sqr	5 µs	6.3 µs	-21.1%
❌	simd_sqr	5.4 µs	6.8 µs	-20.88%
❌	wrm_overhead	1.7 µs	4.4 µs	-59.99%
⚡	block3	190.2 µs	164.8 µs	+15.43%
⚡	block4	233.1 µs	198.2 µs	+17.63%
⚡	bits_20_fixed	183.9 ms	156.6 ms	+17.41%

[xrvdg]

Needs to be rebased on main since #108 has been merged.