Correcting unicode decode error on Windows

Fix Windows UnicodeDecodeError: read data files as UTF-8 (v0.1.1)

taxonomy.json (pretty-printed with ensure_ascii=False) contains non-ASCII
(e.g. subscripts); loaders used read_text() without an encoding, which fails
on non-UTF-8 default locales (Windows cp1252). All data reads now specify
encoding="utf-8". Adds a locale-independence regression test (runs under
LC_ALL=C) and Windows/macOS to the CI matrix.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>