added precomputation of Artin-Schreier LUP decomposition for qadic char 2 sqrt

[alexey-orlov-math]

• Extracted the LUP decomposition of the Artin-Schreier matrix (used for finding initial square root approximation in characteristic 2) into a separate function _artin_schreier_lup, so it can be computed once and reused across multiple qadic_sqrt calls in the same extension field.
• Added a public API (with underscore to signal "use on your own risk"): _qadic_char2_sqrt_precomp_init, _qadic_char2_sqrt_with_precomp, _qadic_char2_sqrt_precomp_clear
• Added tests comparing the precomputed variant against qadic_sqrt on random squares, and a simple benchmark doing R sqrt in the qadic extension of degree d

sqrt [d = 300, R = 500]: 2.0292 s
sqrt (precomp) [d = 300, R = 500]: 0.5945 s
sqrt [d = 500, R = 500]: 6.2686 s
sqrt (precomp) [d = 500, R = 500]: 1.2923 s
sqrt [d = 1000, R = 500]: 37.0105 s
sqrt (precomp) [d = 1000, R = 500]: 5.3692 s

[fredrik-johansson]

Looks good, thanks! Nice speedup.