oj_parse_xml_time parses incorrectly unless there are exactly 9 fractional digits

[wintonpc]

Version 3.16.0. The code assumes exactly 9 fractional digits and miscalculates the nanoseconds otherwise. It also adds 500 picoseconds for a reason I don't understand.

Actual behavior

irb> Oj.load('{"^t":"2023-08-30T14:47:45.732-04:00"}', mode: :object).iso8601(12)
"2023-08-30T14:47:45.000000732499-04:00"
irb> Oj.load('{"^t":"2023-08-30T14:47:45.7327237237-04:00"}', mode: :object).iso8601(12)
"2023-08-30T14:47:52.327237237500-04:00"

Expected behavior

irb> Oj.load('{"^t":"2023-08-30T14:47:45.732-04:00"}', mode: :object).iso8601(12)
"2023-08-30T14:47:45.732000000000-04:00"
irb> Oj.load('{"^t":"2023-08-30T14:47:45.7327237237-04:00"}', mode: :object).iso8601(12)
"2023-08-30T14:47:45.732723723700-04:00"

Generally, I would expect the parsed time to == the result of Ruby's Time.parse.

[wintonpc]

Looks like the 500 picosecond problem was partially fixed in #810