In #98917 we introduced prepared_statements_cache_size, which automatically deallocate LRU prepared statements. In #99663 we (will) introduce pausable portals, which enable executing prepared stmts in an interleaving manner. We need to sort out the expected interaction of both features.

There're several options:

1. Allowing both, meaning when deallocating LRU prepared stmts, we close the associated portals.
2. Disallow setting prepared_statements_cache_size if there're any active portals, or if the session var multiple_active_portals_enabled has been set to true. And vice versa.

Jira issue: CRDB-26158

Epic CRDB-25183