Fix plural forms for locales like Brazilian Portuguese

[cpruijsen]

ngettext now looks up a locale-specific plural form before the
language code, so pt_BR uses n > 1 instead of Portuguese n != 1.
The old path used the language code only (pt_BR → pt), so count 0
took the plural msgstr. Fixes #70.

Lookup is the normalized full locale (pt-BR / pt_BR → pt_br), then
Gettext.getLanguageCode as before. lib/plurals.js gains a pt_br
entry with n > 1. pt is unchanged.

Hyphen and underscore are both accepted because this repo's README
uses BCP 47 hyphens (sv-SE) while gettext and the issue use pt_BR.

The formulas match the issue and GNU gettext's plural table
(pt → n != 1, pt_BR → n > 1), and the Translation Project
pt_BR.po header (Plural-Forms: nplurals=2; plural=(n > 1);).
GNU's table has no other regional exception.

Decision

Locale-then-language lookup, with pt_br as n > 1 and pt left as
n !== 1. Treating pt as Brazilian (CLDR) and adding pt_PT for
European Portuguese would change existing pt callers. Happy to switch
to the CLDR mapping if you prefer.

Did not evaluate headers['plural-forms'] from catalogs. The library
already ignores that header, and compiling it is the GHSA-g974-hxvm-x689
shape (#72).

Test plan

• npx mocha test/gettext-test.js
  - 31 / 32: 0 and 1 → singular, 2 → plural
  - 33: 0 → plural, 1 → singular, 2 → plural
  - 34 still falls back to 35 (36)
• npx grunt (jshint + mocha)