Implement grouping-based exponentiation

[Rjected]

Since we're using this to solve timelock puzzles, the exponent is going to be very large. This can cause issues because the number, when in gmp's data representation, can be gigabytes in size. However, we know that the exponent will be 2^t, where t is the delay parameter.

Since we know t, we should, instead of plugging in 2^t directly, run an exponentiation 2^n times, where n is a parameter. This should be set according to the amount of memory a single instance should take up for its exponent, say 1024 or so. Then each instance will be run t/n times, with the final exponentiation being t % n squarings.

This might re-use a lot of code from kernel_powm_odd, or use fixed_window_powm_odd or sliding_window_powm_odd.