The PDF format and XMP metadata specs define different syntax for dates,
so account for these discrepancies by more carefully constructing the
final timestamps by post-processing strftime() output.
(Tested with Python 3.11.3 on Arch Linux.)
Without passing a tzinfo object to `datetime.now()`, a "naive" datetime
object is created, which is not timezone-aware. To fix the default
date/time detection for non-UTC local timezones, pass
`datetime.timezone.utc` to convert the value to UTC and make the
datetime object "aware".
Also, adjust the strftime() wrappers to use the UTC offsets instead of a
literal `Z`; using the literal `Z` at the end appears to be valid for
ISO 8601, but for some reason it does not successfully convert, whereas
the `%z` placeholder substitutes the UTC offset and successfully
converts.