Fix instance() ignoring the tz argument for aware datetimes

[developer-rpai]

Fixes #841.

pendulum.instance(dt, tz="EST") silently ignored the tz argument whenever dt already carried a timezone: DateTime.instance() unconditionally preferred dt.tzinfo, so no conversion ever happened and the returned value kept the original timezone.

This makes an explicitly passed timezone take precedence:

• Aware datetimes are converted to the given timezone (the instant is preserved).
• Naive datetimes are interpreted in the given timezone, as before.
• When no timezone is passed, behavior is unchanged: aware datetimes keep their timezone and naive ones default to UTC. To make that distinction possible, the tz parameter of instance() (and of DateTime.instance() / Time.instance()) now defaults to None.

Two follow-on adjustments were needed:

• DateTime.utcfromtimestamp(), DateTime.fromordinal() and DateTime.combine() used to pass tz=None through instance() to get a naive result, so they now build their result directly. Their behavior is unchanged (a tzinfo given to combine() now wins over the time's own tzinfo, matching the standard library).
• The parser forwarded its tz option to instance() with a UTC default, which would have turned an unspecified option into an explicit conversion request; it now forwards the option only when one was actually given.

Tests: the full suite passes (1875 passed, 3 skipped), including new regression tests in tests/test_main.py covering conversion for both pendulum.instance() and DateTime.instance(), the reporter's pendulum DateTime case, and naive datetimes with an explicit timezone.

[Wkhanex]

I checked the changed DateTime.instance method with a datetime adapter and ZoneInfo conversion hooks. Four conversions covering both folds of New York’s 2024 fall-back hour to UTC and Brussels preserved the original timestamp. Default-aware input retained its timezone, explicit-zone naive input retained its wall time, and default-naive input used UTC. Those controls support the distinction between converting aware inputs and interpreting naive inputs. Consider a DST-fold regression alongside the new ordinary-date tests. I ran these checks on Python 3.12; isolated method checks, not the full Pendulum parser/timezone suite.