add ability for non-lazy reprojection to write to a sharded fits

[maxwest-uw]

should support #622

If we want to ensure that reprojection can be handled by one function call, we should add the functionality to have a parallel reprojection output the result as a sharded fits. Right now, this functionality is only available to an already sharded fits loaded with lazy=True, but for certain work flows we shouldn't have to write out an intermediary WorkUnit just to run reprojection. It is still worth it imo to write the final result to a sharded WorkUnit instead of pickling+unpickling it in serial after reprojection, as reading it from disk after reprojection is significantly faster.