Find a way to exploit periodicity of behaviours/time functions

[cgibbard]

It seems that for purposes of efficiency, it would be appropriate to exploit known periodicity to be able to cache expensive time functions.

In order to do this, you would likely need to be able to fix a time resolution at which the time function was to be sampled, however, there might be a reasonable way to propagate demand for time resolution, similar to how real arithmetic libraries propagate demand for precision (in order to not lose composability).