Excessive overhead using named tuples and other row cursors

[dvarrazzo]

Row factories using the fields names (such as dict_row, namedtuple_row) use cur.description to get the fields name. This call has a heavy overhead, because it does much more than getting the fields names.

These calls can be optimized. See image for the gains obtained.

Note: this is a pretty tight micro-benchmark, not an indication of the scale of the speedup improvement in a complete program.