Lint tests/ with muff-check; fix the lint violations

[jsh9]

Summary

Follow-up to #71: muff-check now lints tests/ too, and all ~1,060 violations there are fixed.

Config (muff.toml)

• Remove lint.exclude = ["tests/**"].
• Add test-only ignores to the existing "tests/*" per-file ignores, next to PLC2701/PLR2004:
  • INP001: tests/ is not a package; pytest runs the files directly
  • PLR0904: a test class can group many tests
  • PLR6301: test methods in unittest.TestCase classes don't use self (221 hits after the assertion changes below)
  • SLF001: tests can check private members (75 hits)

Fixes in tests/*.py

• PT009/PT027 (476 hits): self.assertTrue/Equal/... → plain assert, and self.assertRaises(Regex) → pytest.raises(match=...), mostly with muff's fixes. The auto-fixer re-spelled some number literals (e.g., 6.4e7 → 64000000.0), so I restored the original spellings. The 40 assertAlmostEqual calls became pytest.approx(b, abs=delta). Where no delta was given, I used abs=1e-7, close to its default of 7 decimal places. The test classes still inherit from unittest.TestCase, so the if __name__ == '__main__' runners still work.
• PTH118/PTH120 (184 hits): f_dir = Path(__file__).resolve().parent / 'files'. Paths passed to PySeismoSoil are wrapped in str(...), because the library checks isinstance(data, str) to tell a file name from an array. np.genfromtxt takes the Path directly.
• RUF043: literal match= strings that contain . etc. are wrapped in re.escape().
• ANN*: -> None on test methods; annotated the nearly_identical() helper and the __init__ overrides.
• The rest, by hand:
  • imports moved to the top of the file (PLC0415)
  • unused variables removed or renamed to _ (F841, RUF059)
  • long comments and strings wrapped (E501), and three over-long test names shortened
  • one statement per pytest.raises() block (PT012)
  • a list comprehension (PERF401)
  • return not r[1, 0] < thres (SIM103; same result as before, including when r is NaN)
  • in test_Vs_cap_is_user_defined, compare with a Vs_cap variable instead of the float literal 1234.5 (RUF069); still an exact comparison
• NPY002: kept as inline # noqa in one SVM test. That test checks on purpose that get_randomized_profile() doesn't change numpy's legacy global random state, so it has to call np.random.seed() and np.random.random().

Docs

• CONTRIBUTING.md and CHANGELOG.md are updated: tests/ is now linted, and tests use plain assert and pytest.raises().

Notebooks

• examples/*.ipynb re-run, as every PR requires.

Test logic is unchanged

I compared every test function with main after parsing both versions. I mapped each self.assert*() call to its plain-assert form and os.path.join to pathlib, and ignored annotations and moved imports. The other differences fall into these groups:

• Equivalent:
  • operand order (0 == x → x == 0)
  • unused variables dropped or renamed to _
  • the list comprehension
  • the Vs_cap variable
  • not r < thres
• Slightly stricter:
  • two pytest.raises() blocks now hold only the call that should raise; the setup moved out (PT012)
  • 11 match= patterns are escaped, so . matches only a literal period
• Slightly looser: the assertAlmostEqual calls without delta now allow an absolute difference of up to 1e-7. unittest's default (rounding to 7 decimal places) is about 5e-8.

Number literal values, comments (apart from line wrapping and the removed # noqa: F841), class bases and module-level code are all unchanged. Both versions collect the same 262 tests, apart from the 3 renamed ones.

Test plan

• pytest tests: 262 passed (same tests as before)
• pre-commit run -a: all hooks pass, including muff-check on tests/
• tox -e run-notebooks: all notebooks ran without errors

🤖 Generated with Claude Code

https://claude.ai/code/session_01TaAeTVJL3sqwsptnDto2WD