Match the Postgres repr in Time and TimeWithTimeZone

[workingjubilee]

By using a Rust-side repr that more closely reflects the internal implementation of Postgres, we head off various issues we potentially could encounter with not being able to reflect the full range that Postgres can in its implementation of these types, versus the edges that time types runs up against (e.g. not having the concept of a temporal "infinity" or "-infinity"). These last two types do not have quite the same problems per se, but we previously were discarding information to achieve a goal in serialization, when that can be done by simply altering the way the type is serialized.

This completes #618 for all of the Postgres temporal types, reducing the likelihood of a forced breakage in the future if the time crate alters its implementation. We also avoid the overhead of calling potentially costly conversion functions that was noted in #504 until such points as serialization.