SQUFOF for large u128 factorization is very slow compared to SIQS

In https://github.com/Pr0methean/FactorBot/blob/7f49a6a899357e8b6bdd9084b25c9b75b49f358e/src/algebraic.rs#L2271, I use yamaquasi's implementation of SIQS to factories numbers between 1u128<<85 and u128::MAX, because the SQUFOF implementation provided by num-prime is much slower.

A good benchmark to reproduce this issue would probably be 319014718988379810428474270189615055511, because it is 13835058055282163729 (the next prime after 3u128<<62) * 23058430092136939559 (the next prime after 5u128<<62). Another option is 0xfffffffffffffffff33333332bdf7d55, which is 0xccccccccccccaa27 * 0x14000000000003623 and was found by ChatGPT as an estimate of the largest u128 biprime that has a smaller factor above 3u128<<62 and a larger factor above 5u128<<62. If you also want to benchmark on a non-adversarial random distribution, try thread_rng().gen() | ((1u128 << 127) + 1) as input.

Start by reproducing the comparison from FactorBot's src/algebraic.rs#L2271 using the two supplied semiprimes and a random u128 input. Inspect num-prime's SQUFOF implementation and compare its timings with yamaquasi's SIQS. Done means large u128 inputs factor substantially faster without breaking existing factorization behavior.