Changes to date->timestamp

I've just committed changes to the main tree which fixes date to
timestamp conversion around daylight savings time boundaries. A
previously-uncalled system routine, mktime(), is used for this
conversion, so we need to keep an eye out for portability issues.

I've added explicit regression tests for the date, time, and time with
time zone data types.

I also uncovered a formatting bug when printing a zero time interval
using the ISO format, and have fixed it. The bug led to three zeros in
the minutes field of the interval.

All regression tests pass on my Linux box (except for the usual
geometric type rounding errors).

- Thomas