Change our `map_partitions` calls to use dask's new `required_columns` arguments

[smcguire-cmu]

We requested this feature in dask to allow our users to have column projection through our custom functions like filters. We should update our code to use this new feature.

Places that we could use it:

• Search Filters
  • this would probably involve having each filter specify which columns it needs to perform the filter
• Crossmatch and Join
  • we would need to switch from delayed to map_partitions for this. Related to Explore dask expressions support using ddf.partitions and map_partitions instead of delayed #300
• NestedPandas calls
  • Any calls such as reduce that we know only need certain columns

There could be other places that we use map_partitions too.