sql/catalog: type hydration under locked leasing may return an error during concurrent DROP

**Describe the problem**
Under locked leasing, type hydration resolves a descriptor at the lease timestamp but resolves its referenced descriptors (the type and its parent schema/database) at the newer read timestamp. When one of those referenced descriptors is deleted in that window by a concurrent `DROP ... CASCADE`, hydration fails and the query surfaces an error such as "unknown schema" instead of transparently completing.

This issue was first seen in https://github.com/cockroachdb/cockroach/issues/173211. `TestLeasedDescriptorTypeHydration` flaked with error ```pq: unknown schema "[104]"``` while a `DROP SCHEMA ... CASCADE `runs concurrently with queries that hydrate descriptors referencing types in that schema.

**Potential Solution**
The hydration path can force a transaction retry when a referenced descriptor is not found at the read timestamp, but the retry must be bounded to avoid retrying for ever. Make sure the solution is valid for all isolation levels.

Jira issue: CRDB-67835

Epic CRDB-65516

Start with TestLeasedDescriptorTypeHydration and trace the locked-leasing type hydration path that resolves referenced descriptors at the newer read timestamp. Reproduce the concurrent DROP ... CASCADE failure, then verify that missing referenced descriptors trigger a bounded transaction retry across all isolation levels without surfacing an unknown-schema error.

Written by the indexing model from the issue text.