sql.py: DbConn: Only cache exceptions for inserts since they are not idempotent, but an invalid insert will always be invalid. If a cached result in an exception, re-raise it in a separate method other than the constructor to ensure that the cursor object is still created, and that its query instance var is set.

sql.py: insert(): Cache insert queries by default. This works because any DuplicateKeyException, etc. would be cached as well. This saves many inserts for rows that we already know are in the database.

sql.py: DbConn.run_query(): Cache exceptions raised by queries as well

sql.py: DbConn.run_query(): When debug logging, label queries with their cache status (hit/miss/non-cacheable)

sql.py: DbConn.run_query(): Also debug-log queries that produce exceptions

sql.py: DbConn: Allow creator to provide a log function to call on debug messages, instead of using stderr directly

bin/map: Pass debug mode to DbConn so that SQL query debugging works again

sql.py: DbConn: DbCursor: Fixed bug where caching was always turned on, by passing the cacheable setting to it from run_query(). Turned caching back on (uncommented it) since it's now working.

bin/map: map_rows()/map_table(): Pass kw_args to process_rows() so rows_start can be specified when using them. DB inputs: Skip the pre-start rows in the SQL query itself, so that they don't need to be iterated over by the cursor in the main loop.

bin/map: Fixed bug introduced in r1718 where the row # would not be incremented if i < start, causing an semi-infinite loop that only ended when the input rows were exhausted. process_rows(): Added optional rows_start parameter to use if the input rows already have the pre-start rows skipped.