Release v19.62.3

Changed

• SQL storage records which physical database each event store, namespace and read-model set uses, in a new DatabaseNames table in the cluster database, and never moves a database once it is recorded. Existing deployments with short names keep their databases unchanged. An existing deployment whose PostgreSQL names were already truncated keeps the shared database where ownership is clear; where data could be stranded, the kernel stops with DatabaseOwnershipUnresolved and the SQL database names documentation describes how to move the data and record the binding. A configured cluster database name longer than the provider limit now fails startup instead of being truncated. (#4849)

Fixed

• On PostgreSQL, an event store and its namespace no longer share one database when their names are longer than PostgreSQL's 63-byte limit. Long database names are now a readable prefix plus a hash and stay distinct, so every namespace migration runs and observers no longer fail with relation "FailedPartitions" does not exist. On SQL Server, names longer than 128 characters are created the same way instead of failing. (#4849)
• A migration that reports "already exists" while migrations are still pending now fails with DatabaseMigrationIncomplete instead of marking the database as migrated. (#4849)