Switch to ducc.misc.special_add_at() if available

This should improve performance quite significantly. Since ducc0 is an optional dependency (and since the function is only available on the unreleased ducc0 branch so far), the switch is only made if the functionality is actually found.

@jroth if you have any relevant real-world benchmarks, I'd be happy about feedback!

@parras I'm not yet happy with how I import this in correlated_fields.py, but I don't have a good idea how to do it better. Any ideas?