Performance improvement for libpq: avoid calling malloc separately

for each field of each tuple.  Makes more difference than you'd think...

Author: tglsfdc