sql.py: mk_insert_select(): INSERT IGNORE: Implemented simple cursor loop that just inserts each row, without ignoring duplicate keys

sql.py: mk_insert_select(): Added ignore param to do an INSERT IGNORE equivalent operation. Added basic structure for INSERT IGNORE. Support embeddable queries without RETURNING columns (which is the case for most INSERT IGNORE queries).

sql.py: run_query_into(): Always set into.schema to None, so that table is created in the temp schema in debug_temp mode

sql.py: DbConn.DbCursor.execute(): Don't cache ADD COLUMN queries because the rest of the query must be unique in order to use caching

sql.py: add_col(): Don't ignore already-existing columns because sometimes name truncation causes collisions, requiring the caller to version the name. add_index_col(): Version the column name to avoid collisions.

sql.py: add_index(): Version index names to avoid collisions

sql.py: DbConn.DbCursor.execute(): Always cache structural changes, not just if they throw an exception

sql.py: add_index(): Fixed bug where expr could not be deep-copied until ensure_not_null() had been run on it, because ensure_not_null() modifies the index_cols of the table and this needs to be modified on the original table

sql.py: add_index(): Fixed bug where expr could not be deep-copied until ensure_not_null() had been run on it, because ensure_not_null() modifies the index_cols of the table and this needs to be modified on the original table

sql_gen.py: EnsureNotNull: Added to_str() which uses the column's index col if available. This works around a PostgreSQL bug where indexes on expressions are sometimes not used, but indexes on plain columns are used.