why: Only the "latin" subset was downloaded. Characters outside basic
Latin (accented letters like ñ, ř, ž common in contributor names and
code comments) fell back to system fonts, breaking the typographic
consistency of IBM Plex.

what:
- Add _UNICODE_RANGES dict mapping Fontsource subset names to CSS
  unicode-range descriptors for latin, latin-ext, cyrillic,
  cyrillic-ext, greek, and vietnamese
- Add _unicode_range() lookup function with empty-string fallback
  for unknown subsets (omitting the descriptor = all codepoints)
- Support "subsets" (list) config key alongside legacy "subset" (str)
  in _on_builder_inited — iterates all subsets in the download loop
- Emit unicode-range in font_faces dict; template renders it in
  @font-face blocks so browsers only download subset files when
  characters from that range appear on the page (zero cost for pages
  that only use ASCII)
- Preload uses the first (primary) subset only — typically "latin"
- Update page.html template: conditional unicode-range line in
  @font-face blocks
- Add 8 tests: _unicode_range (4), multi-subset builder-inited (2),
  legacy subset backward compat (1), preload primary subset (1)

why: Accented characters (ñ, ř, ž, ö) in contributor names, docstrings,
and code comments fell back to system fonts. latin-ext covers U+0100-02FF
and extended Latin ranges used across European languages.

what:
- Change "subset": "latin" to "subsets": ["latin", "latin-ext"] for both
  IBM Plex Sans and IBM Plex Mono
- Total font files: 32 (2 fonts × 2 subsets × 4 weights × 2 styles)
- Zero performance cost for ASCII-only pages: unicode-range descriptors
  tell the browser to skip latin-ext downloads when no extended chars
  appear on the page