Builtin prime table up to 20 bits; ulong_extras cleanup

[fredrik-johansson]

The static bit table of odd primes up to 2^15 is complemented with a static bit table mod 30 up to 2^20 (costing 35 KB), making n_is_prime nearly instantaneous up to 20 bits. We also search these tables directly in n_nextprime, significantly speeding up generating primes up to one million.

Timings for n_is_prime, average for 10^5 random inputs of the given bit size:

bits     old       new     speedup
 1   | 5.66e-09  2.49e-09   2.273x
 2   | 1.02e-08  2.50e-09   4.080x
 3   | 1.06e-08  2.49e-09   4.257x
 4   | 1.13e-08  2.50e-09   4.520x
 5   | 1.29e-08  2.50e-09   5.160x
 6   | 1.49e-08  2.51e-09   5.936x
 7   | 1.61e-08  2.49e-09   6.466x
 8   | 1.87e-08  2.68e-09   6.978x
 9   | 2.09e-08  2.54e-09   8.228x
10   | 2.41e-08  2.75e-09   8.764x
11   | 1.32e-08  2.64e-09   5.000x
12   | 1.35e-08  2.71e-09   4.982x
13   | 1.43e-08  2.73e-09   5.238x
14   | 1.53e-08  2.76e-09   5.543x
15   | 1.52e-08  2.82e-09   5.390x
16   | 7.01e-08  5.22e-09  13.429x
17   | 8.22e-08  5.39e-09  15.250x
18   | 9.50e-08  5.59e-09  16.995x
19   | 1.08e-07  5.76e-09  18.750x
20   | 1.21e-07  5.95e-09  20.336x
21   | 1.35e-07  1.35e-07   1.000x
22   | 1.47e-07  1.47e-07   1.000x

Timings generating the primes up to 10^3 sequentially using repeated calls to n_nextprime:

old:   1.53e-06
new:   6.02e-07  (2.5x speedup)

Timings generating the primes up to 10^6:

old:   0.00827
new:   0.00057  (14.5x speedup)

For comparison, repeatedly calling n_primes_next takes 3.92e-07 seconds up to 10^3 (where it's just copying from the small primes table) and 0.00115 seconds up to 10^6 (where it's sieving). Obviously, n_primes_next would benefit significantly from using the same tables as n_is_prime and n_nextprime, but I'm not going to touch that function for now (it wants other optimizations too).

This PR also removes the obsolete n_is_oddprime_small and n_is_oddprime_binary functions and replaces the obsolete n_is_probabprime with a wrapper of n_is_prime.