fix(number, lists): correct negative ordinal suffixes and support general iterables with oxford_comma

[miskatul-anwar]

Summary

This PR resolves two key issues and enhances developer ergonomics:

1. Negative Integer Ordinal Suffix Calculation (src/humanize/number.py):

   • Fixed remainder calculations in ordinal() for negative integers (e.g. -21 previously returned '-21th' due to Python's modulo behavior on negative numbers). Using abs(value) ensures correct suffixes (e.g., -1st, -2nd, -3rd, -11th, -21st).

2. Iterable & Oxford Comma Support in natural_list (src/humanize/lists.py):

   • Updated natural_list to accept any general Iterable (such as generators, tuples, sets) instead of strictly requiring list.
   • Added an optional oxford_comma: bool = False parameter to format lists as 'one, two, and three'.

3. Source Import & Test Resilience (src/humanize/__init__.py, tests/):

   • Added a try...except ImportError fallback for _version.py in __init__.py so importing directly from source without build artifacts works smoothly.
   • Added pytest.importorskip for optional test dependencies (freezegun, pytest-codspeed) to prevent test crashes in minimal development environments.

Empirical Test Verification

Ran pytest:

319 passed, 3 skipped in 0.39s

[Mukller]

Verified locally on the PR branch vs release 4.16.0 (Python 3.13):

The negative-ordinal fix is real and correct — Python's modulo makes the old code produce wrong suffixes:

input	release	branch
ordinal(-1)	-1th	-1st
ordinal(-2)	-2th	-2nd
ordinal(-11) / (-111)	-11th / -111th	same (11/12/13 rule preserved via abs % 100)

natural_list improvements verified too: plain generators now work (release raises TypeError), oxford_comma=True yields "a, b, and three"-style output while default stays byte-identical, two-item lists ignore the flag correctly.

pytest tests/test_lists.py tests/test_number.py — 243 passed.

One request before merge, on an unrelated bundled hunk: the __version__ fallback hardcodes "4.12.2.dev0" when _version.py is missing. That misreports the actual version in exactly the broken environments where accurate versions matter most (bug reports from source checkouts would claim 4.12.2). Suggest either dropping that hunk from this PR (it's orthogonal to both fixes here) or using a neutral placeholder via importlib.metadata.version("humanize") with a fallback of "unknown".

Everything else: approve-worthy.

[miskatul-anwar]

Thanks for the review @Mukller! I have dropped the _version.py fallback hunk from this PR as suggested.