consider how to distribute heterobatch wheels

[lgarrison]

With the heterobatch interface (#66), we are calling finufft from C++. This is a departure from the standard interface, which calls finufft from Python. To avoid having multiple finufft libraries at runtime, we might want to update the standard interface to call finufft from C++, too. This could use the process_single_series() functions from the heterobatch helpers.

At the same time, we could consider whether there are opportunities to simplify the heterobatch interface, such as by unifying the chi2 and non-chi2 interfaces. Maybe the standard interface could undergo a similar unification.

For now we are still using cufinufft from Python. So we might end up in a situation where we are using bundled finufft but external cufinufft. Not ideal, but it also should be harmless.