Formatting years prior to 1000 with %F should produce four digits for year, not three.

Timezone claims to format dates "with a full implementation of strftime formats, including the GNU date extensions." The [stftime man page](http://man7.org/linux/man-pages/man3/strftime.3.html) describes `%F` as "equivalent to `%Y-%m-%d` (the ISO 8601 date format)." There is, however, ambiguity here when dealing with years prior to 1000. The man page also describes `%Y` as "the year as a decimal number including the century." But [ISO 8601](http://en.wikipedia.org/wiki/ISO_8601#Years) "prescribes, as a minimum, a four-digit year [YYYY]."

I am ambivalent about whether `%Y` should be changed to always include four digits. Some languages (e.g. ruby) treat `%Y` as a `CCYY` value even for single digit centuries, others as variable width; in python, a century less than 1000 is raised as an error. I believe it is appropriate to see "May 1, 900" as the output for the format string `%b %-e, %Y` on a date of `0900-05-01`, but would already see "May 1, 0900" from ruby anyways.

However, I believe if `%Y` remains variable width, `%F`'s ambiguity should be resolved in the direction of conforming to ISO 8601 rather than conforming to equivalence with `%Y-%m-%d`. `%F` is primarily used as a machine-oriented format string (particularly when generating ISO 8601 strings). Without an ISO 8601 compliant `%F`, there is no format string that can be provided to produce an ISO 8601 output.