Add worst case tests for fixed dlog.

[NicholasPatrick]

big_safe_random

Pohlig-Hellman algorithm can be quite effective in existing tests, even for precomputation. Safe primes provide the worst case for Pohlig-Hellman. Example hack, note the slightly longer precompute time of 14ms compared to 12ms in sample in stderr.

big_safe_random_g2

Some solutions try to solve for discrete logarithm of $k$ via a reduction to $k, gk, g^2k, ...$ until one is "easy enough" to solve (e.g. when it's smooth or $k/p$ is very close to a fraction with a small denominator), treating this sequence as random. Such solutions will be slowed down when $g=2$ because if a particular value isn't suitable, twice that value is likely to not be suitable either.

Example of a solution being slower when $g=2$ because:

• the solution tests for smoothness
• the solution tests for closeness to a fraction with small denominator (note the even longer precompute time of 16ms in the hack)

[maspypy]

I'll merge it, thank you!