[tamnd]

Summary

pg_regress starts each run with ALTER DATABASE "regression" SET lc_messages TO 'C' and five more settings. rupg gave "ALTER is not supported yet", so pg_regress stopped before the first test and the pass count was 0 of 239. This PR runs ALTER DATABASE ... SET and applies the stored settings when a session of the database starts.

Changes

• ALTER DATABASE name SET, SET ... FROM CURRENT, SET ... TO DEFAULT, RESET name and RESET ALL run as AlterDatabaseSet and AlterSetting do. The checks of validate_option_array_item come first: the name, the context and the value. A new value goes in place of the value of the same parameter, as GUCArrayAdd does.
• The store keeps the settings of each database. DROP DATABASE removes them.
• A new session applies the settings with the source database before the values of the startup packet, as process_settings does. RESET goes back to them, and pg_settings.source shows database.
• pg_db_role_setting shows a row for each database with settings.
• The statement runs in the simple and the extended protocol.

Known differences

• The change does not wait for the end of the transaction. A rollback does not undo it.
• ALTER ROLE ... SET is not supported yet, because rupg has one role.
• A stored setting that is no longer valid gives no warning at the start of a session.

Tests

• connection/alter_database_set.test is the output of PostgreSQL 19 for the same script: the errors, the placeholders and the rows of pg_db_role_setting.
• database_settings starts a session of the database and checks the value, the source and RESET.

fmt, clippy, the workspace tests and the checks of xtask pass on server1.