CAST Timestamp -> String doesn't properly handle timezones with sub-minute offsets #19792
Description
One of the timestamp -> string test cases is -62135596800000000 microseconds from the unix epoch, or 01/01/0001 00:00:00 GMT
Technically the timezone offset at this time in America/Los_Angeles is -07:52:58. This causes the following error:
Expected: ARRAY<STRUCT<ColA STRING>>[{"0000-12-31 16:08:00-07:52"}]
Actual: ARRAY<STRUCT<ColA STRING>>[{"0000-12-31 16:07:02-07:52"}]
Note that ZetaSQL expects us to completely truncate the second part of the offset. It's not used when subtracting from the origin datetime, and it's not included in the offset string. However when we perform this conversion, joda time uses the second part of the offset, and thus our time string is off by 58 seconds.
Imported from Jira BEAM-8073. Original Jira may contain additional context.
Reported by: amaliujia.