Implement circular buffer for History array

[cglosser]

The History class currently uses an array of length num_timesteps in its time dimension which will not scale for exceptionally long simulations. This should probably be reworked to use a circular buffer so that new timesteps overwrite old ones and the array does not grow inordinately. The circular buffer will have to be at least as long as the maximum number of timesteps required for a signal to traverse the system, however (r_max / (c * delta_t)).

[tbertus]

As of now, the range of the time dimension of the history array is -window, num_timesteps). With a circular buffer it no longer strikes me as necessary to have the time dimension with -window based indices. Would you agree?

[tbertus]

The circular buffer works, but only when it is above a seemingly arbitrary size. Below that size I get a disagreement of ~1e-13 between the results of the new and stable builds. I am sure to keep enough time steps to account for the max traversal time. If you have any debugging advice, I'd appreciate it!

[cglosser]

That's really odd - it's such a small difference, but it should be exactly zero. It might be the max traversal time + some number of tail timesteps for the interpolation; what happens if you make it r_max / (c * dt) + interpolation_order?