Aggregation control still non-optional

[aulemahal]

Almost the same issue as #549. PR #551 fixed the pydantic model issue, however the code itself expected an aggregation_control object to exist, so the bug remains.

First, I'm not sure if aggregation_control should be optional. When it is None, the keys of the dataset are simply all the columns joined together, including the path. There is also no variable_column_name and thus no derived variable registry and ESMDataSource can't select the expected variable.

I have a version where (AFAIK) intake_esm now works with aggregation_control=None, but I think an alternative would be to make it required. What do people think?