fix: correctly handle negative UTC offsets in isoformat

[binggao1230]

• This contribution adheres to CONTRIBUTING.md.

What does this Pull Request accomplish?

Fixes #52.

hightime.datetime.isoformat() built its result by splitting the base datetime.isoformat() string on "+" to separate the UTC offset, then re-joining with "+". Negative offsets use "-", so the split found nothing and the high-precision fractional seconds were appended after the offset instead of after the seconds:

>>> ht.datetime(2025, 1, 1, yoctosecond=40, tzinfo=dt.timezone(dt.timedelta(hours=-1))).isoformat()
'2025-01-01T00:00:00-01:00.000000000000000000000040'

This replaces the split with a regex match for the offset suffix ([+-]HH:MM or [+-]HH:MM:SS) at the end of the string, so the fractional part is inserted after the seconds and the offset is re-attached last. It also removes the xfail from the existing negative-offset case in tests/test_datetime.py, which now passes.

Why should this Pull Request be merged?

The output was not a valid ISO 8601 string and could not be parsed back, for any sub-microsecond datetime west of UTC. Positive offsets and naive datetimes were unaffected, which is why it went unnoticed. The repository already had a test for the case, marked xfail against #52.

What testing has been done?

• python -m pytest tests/test_datetime.py -k isoformat — 12 passed; the negative-offset case fails on master with the xfail removed, confirming the test covers the fix
• python -m pytest tests/ — 408 passed
• ni-python-styleguide lint — clean
• mypy — Success: no issues found in 8 source files
• CI on this PR: 29/29 checks passing

[bkeryan]

@gaoflow, thank you for contributing.

Please use the project's pull request template, like you did in #136

[binggao1230]

Sorry about that — description rewritten using the template, same as #136. Rechecked on current master while I was in there: 408 tests pass, lint and mypy clean, 29/29 CI green.