Fix KeyError in diff_for_humans for the zh locale

[dylanpulver]

Pull Request Check List

• Added tests for changed code.
• Updated documentation for changed code.

---

diff_for_humans() raises KeyError: 'time' for the zh locale whenever it is given a
second datetime (the non-relative-to-now form), for every unit:

>>> a = pendulum.datetime(2024, 6, 15, 12, 0, 5)
>>> b = pendulum.datetime(2024, 6, 15, 12, 0, 0)
>>> a.diff_for_humans(b, locale="zh")
KeyError: 'time'
>>> a.diff_for_humans(b, locale="ja")
'数秒 後'

zh/custom.py:9-10 holds "after": "{time}后" / "before": "{time}前" — named placeholders —
while DifferenceFormatter.format() substitutes positionally
(difference_formatter.py:115 and :159, both locale.get(key).format(time)). All 28 other
locales use {0}, so this changes only the zh data; no code change.

Base is master per the template note (bug fix, not a behaviour change).

Why nothing caught it: the relative-to-now form reads translations.relative.* (the
generated CLDR table) and never touches these two custom entries, and there is no
tests/localization/test_zh.py — zh, en* and pt_br are the locales with no test file.
Added one for zh, modelled on test_ja.py, covering both forms.

How it was found, since it's a reasonable question: a sweep, not a bug report from real
use. I called diff_for_humans across all 29 locales x 8 durations x both signs x
absolute on/off (928 calls) and looked for exceptions or unsubstituted {. zh was the
only locale that failed, 16/32 of its calls — exactly the is_now=False, absolute=False path.
{time} appears nowhere else in src/ or tests/.

Measured: pytest tests 1840 passed / 3 skipped before, 1841 / 3 after (same command,
same venv). Mutants: reverting the two strings and keeping the new test re-raises all 16
sweep failures and fails test_zh.py.

Honest limitation: a second mutant — keeping {time} and making the formatter call
.format(time, time=time) — also passes the sweep and the new test. So the test does not by
itself pick between the two remedies. I chose the data fix because the alternative sanctions
two placeholder conventions in the locale files; if you would rather the formatter be
tolerant, say so and I'll switch it.

Not tested: I did not add test files for en, en_gb, en_us or pt_br, which also
have none. ruff format --check reports a pre-existing blank-line diff in 28 of the 29
locales/*/custom.py files including this one; I left it alone.

Disclosure: the diff, the test and this description were drafted with Claude Opus 4.8.