test: pin invalid timezone error on EncounterRecencyTransformer

[HarshRajSinghania]

Closes #204.

Summary

Adds a regression test so an invalid timezone= on EncounterRecencyTransformer raises KeyError that names the zone. The pre-#202 fallback turned a bad zone into TypeError("Already tz-aware"); this test would fail if that path returned.

No production code change. __init__ still does not validate timezone (sklearn convention noted in the issue).

Motivation

#204: tests/test_encounter_timezone.py only used valid timezones or none. The behaviour restored in #202 was unguarded.

Implementation

• Reuse the two-row fixture from test_encounter_recency_localises_to_a_non_utc_timezone.
• Construct the transformer with timezone="Not/AZone" and call fit_transform.
• Assert pytest.raises(KeyError, match="Not/AZone") so both pandas 2 (pytz) and pandas 3 (zoneinfo) are covered without importing pytz.
• Assert the message does not contain Already tz-aware.
• CHANGELOG.md [Unreleased] entry and CONTRIBUTORS.md line, as required by the project.

Testing

Intended local commands from the issue:

python -m pip install -e ".[dev]"
python -m pytest tests/test_encounter_timezone.py -q
make ci

This environment could not install scikit-learn (PyPI 502), so the new test was not executed here. CI on this PR should run the file and the full matrix.

[shivamlalakiya]

Closing this as a duplicate: #204 was claimed by @Kaap10 on 2026-09-12 and confirmed on the issue the same day, and #221 has now merged as dba8cd5.

That is the claim-before-PR rule from #187 working as intended, and it is on me that it is easy to miss: the claim is a comment on the issue rather than an assignee, so the issue looks unclaimed from the list view. Checking the issue thread before starting is the only reliable way right now.

Nothing wrong with the test itself. If you want another, #50, #53 and #55 are all open with zero comments, so no claim to collide with:

• test: PropensityScorer.predict_proba never exercises its single-class branch #50 PropensityScorer.predict_proba single-class branch
• test: cli's _score_array never exercises the plain predict_proba fallback branch #53 cli _score_array plain predict_proba fallback branch
• test: FiscalYearGroupedSplitter.split never exercises the no-prior-training-data skip branch #55 FiscalYearGroupedSplitter.split no-prior-training-data skip branch

Thanks for #216 and for this one.