[parijat1222q]

Normalize n with Math.abs once before evaluating generated JavaScript plural rules. This matches Python's extract_operands and the Unicode definition of plural operands as absolute values.

For {'one': 'n is 1', 'few': 'n in 2..4'}, Python selects one for -1 and few for -2, while the current generated JavaScript selects other for both.

Update the generated-source expectations and add always-on normalization checks for empty, direct, integer, modulo and range rules. Four execution tests compare the generated function with Python for negative integers/fractions, zero and positive controls. They use Node.js when available and skip only those execution tests otherwise, without adding a required dependency. All four execution regressions fail on unchanged master.

Validation against 9a3ef47, Linux, imported CLDR 48.2:

• All 19 configured Tox environments pass: CPython 3.8-3.15, PyPy 3.10, and the listed pytz/setuptools/Jinja variants, including coverage. CPython 3.15: 7,924 passed, 7 skipped, 2 expected failures. PyPy: 7,923 passed, 7 skipped, 3 expected failures.
• Plural tests: 47 passed, including the Node.js 22.23.3 execution regressions.
• Configured, pinned pre-commit hooks on all files, source/wheel build, and git diff --check pass.
• Separate generated-JavaScript/Python comparison: 29,795 integer checks across 59 distinct cardinal/ordinal rules pass. Rules referencing existing unsupported compact-exponent operands were excluded from that extra comparison; they are not changed here.

Test runs used process-local TZ=UTC: two unrelated host-timezone assertions fail with Asia/Calcutta on both unchanged master and this branch. No tests were deselected. The existing pytest parametrization warning remains. Windows/macOS were not run locally; upstream CI is still pending.

Fixes #1413