UDT deserialization: default to attribute-accessible rows, and add a `register_user_type`-equivalent

Background: in cassandra-driver, UDT columns are never plain dicts. By default, UserType.make_udt_class()
(cassandra/cqltypes.py:976-1010) auto-generates a namedtuple per UDT from its schema (field names come
from CREATE TYPE), giving attribute access (row.location.street) with zero configuration.

On top of that, Cluster.register_user_type(keyspace, udt_name, klass) (cassandra/cluster.py:1639-1692)
lets a user map a UDT to their own class — both directions: binding a query parameter of that UDT
expects an instance of klass, and result rows for that UDT come back as klass instances. It's a global
registry keyed by (keyspace, udt_name), cached and invalidated on re-registration.

Needs to be implemented:

1. Change the default UDT representation from plain dict to something attribute-accessible (a
   namedtuple-like object generated from the UDT's schema, matching the old driver's zero-config
   default), so .field access works out of the box without any registration step.
2. Expose a register_user_type-equivalent — a way to map a given UDT (by keyspace + name) to a
   user-supplied Python class, used both when binding parameters of that UDT and when deserializing
   result rows containing it.

Start with the behavioral references in cassandra/cqltypes.py and cassandra/cluster.py, especially UserType.make_udt_class() and Cluster.register_user_type(). Then locate the corresponding UDT binding and result-row deserialization entry points in this repository. Done means default UDTs support attribute access without registration and a user class is honored in both directions.