Use faster code for remainder by p in qsieve_evaluate_candidate

[fredrik-johansson]

The quadratic sieve spends a decent portion of its time in qsieve_evaluate_candidate which checks the remainder of the candidate i by all factor base primes. There is maybe a better way to implement this function, but a trivial improvement is to replace n_mod2_preinv with a Lemire-Kaser-Kurz remainder. This speeds up the quadratic sieve 5-20%.

Time (single-threaded) to factor nextprime(10^n)*nextprime(10^(n+1)) with fmpz_factor:

         n        old        new   speedup
        10      0.004    0.00402   0.995
        11    0.00131    0.00131   1.000
        12     0.0151     0.0129   1.171
        13     0.0104    0.00886   1.174
        14     0.0107     0.0094   1.138
        15      0.012     0.0105   1.143
        16     0.0192     0.0166   1.157
        17     0.0217     0.0194   1.119
        18     0.0341     0.0283   1.205
        19     0.0317      0.029   1.093
        20      0.046     0.0406   1.133
        21     0.0653     0.0586   1.114
        22     0.0966     0.0879   1.099
        23       0.18      0.161   1.118
        24      0.441      0.383   1.151
        25      0.434      0.402   1.080
        26      0.533      0.456   1.169
        27      0.911      0.795   1.146
        28      1.259      1.112   1.132
        29      2.067      1.927   1.073
        30      2.539      2.269   1.119
        31      7.293      6.666   1.094
        32      7.502      6.671   1.125
        33     14.411     12.894   1.118
        34      23.94     20.594   1.162
        35     41.296      36.51   1.131