sql.py: DbConn: Always set the transaction isolation level to READ COMMITTED so that when a table is locked for update, its contents are frozen at that point rather than earlier. This ensures that no concurrent duplicate keys were inserted between the time the table was snapshotted (at the beginning of the transaction for SERIALIZABLE) and the time it was locked for update.

sql.py: put_table(): Removed locking output tables to prevent concurrent duplicate keys because that is now done automatically by mk_select()

sql.py: mk_select(): Filtering on no match: Lock the joined table in EXCLUSIVE mode to prevent concurrent duplicate keys when used with INSERT SELECT

sql_gen.py: Added underlying_table() and use it in underlying_col()

main Makefile: schemas/rotate: Fixed bug where needed to run schemas/public/install, not full schemas/install, after renaming public schema

sql.py: put_table(): Lock output tables to prevent concurrent duplicate keys

sql.py: Added lock_table()

bin/map: connect_db(): Only use autocommit mode if verbosity > 3, to avoid accidentally activating it if you want debug output in normal import mode

bin/map: connect_db(): Only use autocommit mode if verbosity > 2, because it causes the intermediate tables to be created as permanent tables, which you don't want unless you're actually debugging (verbosity = 2 is normal for column-based import)

sql.py: put_table(): remove_all_rows(): Changed log message to "Ignoring all rows" because NULL is not necessarily the pkey value that will be returned for the rows