fix(linux): strict MySQL saves in the server zone, PostgreSQL data fixes

[datlechin]

• MySQL: the workspace session adds STRICT_ALL_TABLES, so a server started with a lenient sql_mode refuses a value the column can't hold instead of altering it. Editor sessions keep the server's mode.
• Core: DestructiveChange::SetNotNull, so making a column NOT NULL goes through the review dialog.
• MySQL: reads stay in UTC; a save or schema change switches to @@GLOBAL.time_zone and back (a connection that can't switch back is closed). TIMESTAMP values bind through CONVERT_TZ. An instant in a repeated DST hour, which MySQL and MariaDB store an hour off, refuses the batch. TIMESTAMP constant defaults are restated in the server zone. Design in server_zone.rs and docs/drivers/mysql.md.
• PostgreSQL: a narrowing USING casts to the unbounded type (varchar, bpchar, varbit, arrays), so a value that doesn't fit is refused instead of cut.
• PostgreSQL: arrays export as quoted array literals ('{"1","2",NULL}') that insert back into typed array columns.
• PostgreSQL: a SELECT the cursor stopped at the row limit ends as Truncation::ServerStopped, and its banner says the statement did not run past the rows shown.
• Temporal values: a fraction finer than the column keeps is ScaleExceeded; 24:00:00 parses for time-of-day columns.
• Docker tests on a lenient MySQL/MariaDB, a +07:00 server, a repeated hour, narrowing, and array round-trip.
• Tests updated on purpose: unchanged_catalog_type_is_not_type_change, two PostgreSQL DDL tests that expected USING ...::varchar(320), the banner text tests, select_inside_begin_uses_savepoint_cursor_row_limit.