test: prune tautological + redundant tests (stacked on #334)

[thodson-usgs]

Draft, stacked on #334. The first commit is #334 (mock the live NWIS
tests); the net-new change here is the second commit (the prune). The diff
below reduces to just the prune once #334 merges.

What this prunes

Once #334 mocked the live NWIS tests, two classes became tautological — they
assert the columns/shape of the fixture they're handed, with no production logic
between the mock and the assertion that isn't already covered elsewhere:

• nwis_test.py::TestSiteseriesCatalogOutput (Parameter code not returned from site information #34) — RDB column parsing is
  covered by rdb_test.py + TestReadRdb; and since the mock returns the chosen
  fixture regardless of the request, the seriesCatalogOutput param itself is
  never exercised. Deleted (+ the orphaned nwis_site_seriescatalog.txt fixture
  and now-unused imports).
• nwis_test.py::TestTZ (Error when combining Dataframe from stations with different time zones #60) — once mocked it is identical to
  test_iv_service_answer (same nwis_iv_mock.json, same "datetime index"
  assertion); the two [:25]/[:20] variants differ only by a slice the mock
  ignores. Deleted.
• waterdata_filters_test.py::test_construct_filter_passthrough — filter
  is forwarded verbatim, so it's subsumed by the parametrized
  test_construct_filter_on_all_ogc_services (8 services + filter-lang) and by
  the long-filter fan-out tests. Deleted.

Net: −8 tests, no coverage lost.

Whole-suite review

I ran a 4-agent review across all 17 test files (~7,800 lines) with strict,
conservative criteria (tautological / redundant / vacuous / dead). The suite is
well-curated — the deletions above are the only defensible ones. Notable
kept items and why:

• nwis_test.py::TestMetaData (6 near-identical assert md.site_info) — each
  passes a distinct major filter through NWIS_Metadata; kept (only low-confidence
  redundancy).
• waterservices_test.py::*_site_value_types (4) — verify each getter accepts
  both sites="x" and sites=["x"]; a real per-getter input contract.
• waterdata_test.py::test_get_samples_legacy_camelcase_kwargs_warn — the only
  test asserting the DeprecationWarning fires on a real get_samples call
  end-to-end (the mapping-wide tests use a spy / suppress warnings).
• wqp_test.py what_* (8) — identical boilerplate assertions, but each is the
  sole check that a public wrapper maps to the right service string; a
  parametrize-refactor candidate, not a prune.
• chunking / progress / nearest / ratings / nldi / ngwmn — each test pins a
  distinct branch; nothing cut.

Verification

nwis_test.py + waterdata_filters_test.py: 118 passed, offline, ruff/mypy
clean.

🤖 Generated with Claude Code