test: parametrize two clean duplicated-test clusters

[thodson-usgs]

Draft. Follow-up to #336 (the WQP what_* parametrization, merged). Test-only.

Scope (revised after a readability pass)

Parametrization trades named tests for a data table, so it's only a win when it
reduces LOC and the cases are pure data-variations of one assertion with
no per-case rationale lost. Two clusters clear that bar:

• _format_api_dates (waterdata_utils_test.py) — 12 one-line equality tests
  → 1 (13 rows, explicit ids). Big, clean reduction.
• TestNormalizeStrIterable iterable-normalizes (waterdata_test.py) — 4 → 1;
  same assertion, only the container type varies (list / tuple / Series / ndarray).

Net −16 lines, every case preserved, readable ids
(test_format_api_dates[offset_to_utc], …[ndarray]). 89 tests pass; ruff/mypy clean.

Deliberately NOT parametrized

After a second look I dropped three clusters I'd initially collapsed — they made
the tests harder to read:

• _parse_retry_after — these were 4 distinct documented behaviors (None
  fallback, delta-seconds, clamp-negative, unparseable), each with a docstring
  explaining the why. Flattening to bare tuples deleted that rationale.
• _build_filter — the long CQL expected-strings forced multi-line tuples, so
  it didn't reduce LOC and read worse than 5 named one-line tests.
• reject-non-string-types — only 3 cases; the param table added lines, and the
  test names already documented each case.

Conventions used / proposed

• Parametrize only when it cuts LOC and the cases are one-assertion data-variations.
• Keep named tests when each encodes a distinct behavior or carries explanatory docs.
• Always give explicit, readable ids (or pytest.param(..., id=...) for short
  data); keep the body to a single assertion.

🤖 Generated with Claude Code