Experiment: for each duration instance: store all date/time fields as zero-or-greater values and provide an positive/negative attribute

[jayaddison]

Attempt to implement the logic described by: #8 (comment)

Essentially:

• Determine the 'time distance' described by the arguments provided
• Store the positive values required to represent that
• Store a flag to indicate whether the distance was positive or negative

With this in place, it should be possible to update the implementation of timedelta_isoformat.timedelta.isoformat(...) so that all of the designator-separated values it emits are positive.

[jayaddison]

Edge-case / potential conflict with year-and-month support:

Distances defined in terms of years/months are naturally ambiguous -- they only make sense when provided with a reference point and a calendar scheme (to determine leapyear frequencies and magnitude).

If year/month values are present and positive, and if any subtracted date/time elements are small enough to avoid turning the distance negative, then it's ok (albeit not completely unambiguous) to a positive representation of the entire distance, including year-and-month values.

It may be easiest to begin experimentation from a version of the code that does not support non-zero year/month values.