Rewrite timestamp handling to support different representations.

Rewrite timestamp handling to support different representations.

Instead of just allowing sub-second resolution times to be passed
as the fractional part of a floating point value, also support passing
times as an array ref, with separate second and nanosecond portions.
Decided to do this because the floating point representation can lose
precision (it's represented internally as a double). It tends to get
squirrelly around about a tenth of a microsecond.