Optimize `__typename` filtering: narrow queried indices

[myronmarston]

In #1169, we added support for filtering abstract types on their __typename via a _typename filter field. There's an additional query optimization we should implement: we can narrow down the list of searched indices based on the given __typename. For example, consider this type hierarchy:

# The DistributionChannel hierarchy has two branches:
#   DistributionChannel (index: distribution_channels)
#   ├── Wholesale            (interface, inherits distribution_channels)
#   │   ├── DirectWholesaler (concrete, distribution_channels index)
#   │   └── BrokerWholesaler (concrete, distribution_channels index)
#   └── Retail               (interface, inherits distribution_channels)
#       └── Store            (interface, inherits distribution_channels)
#           ├── OnlineStore  (concrete, distribution_channels index)
#           └── PhysicalStore (concrete, physical_stores index)

Given a query like:

distribution_channels(filter: {
  _typename: {equal_to_any_of: ["PhysicalStore"]}
}) {
  # ...
}

...we know that matching documents can only live in the physical_stores index, but, as currently implemented, we'll also query the distribution_channels index. It would be more optimal to only query the physical_stores index.

One thing to bear in mind: supporting this requires proper set algebra so that it works if a _typename filter is under not, any_of, etc. Most of the set algebra logic already exists and can be reused. It's implemented in FilterValueSetExtractor which is used by the RoutingPicker and IndexExpressionBuilder.