Map a DialectId to the upstream sqlparser dialect used to parse it.
MariaDB intentionally returns MySqlDialect. We previously used a
thin MariaDbDialect wrapper, but sqlparser gates many MySQL-only
features (ON UPDATE timestamp column option, table hints, LIMIT a, b
shorthand, etc.) behind dialect_of!(MySqlDialect) macros that
downcast via Any. Any wrapper type, however faithful in trait
forwarding, fails those is::<MySqlDialect>() checks and silently
loses dozens of MySQL-superset features that MariaDB needs. Using
MySqlDialect directly is the only reliable way to get the full
MariaDB grammar parsed today; the application-level distinction is
preserved via this DialectId enum.
MariaDB input is pre-processed by parse::parse to handle the bare
--<EOL> comment form that real mariadb-dump output contains.