yoda: snap monospace cell width via Unicode East Asian Width

Replaces the previous heuristic (font_monospace_em_width.is_none() ⇒ 2
cells) which was reviewed as unsound: Arabic, dingbats, math symbols and
other narrow scripts pulled from non-monospace fallback fonts would all
have been forced to 2 cells. It also didn't handle ZWJ emoji clusters
or ambiguous-width chars correctly.

Proper fix, computed at shape time when `line: &str` is in scope:
- new ShapeGlyph.terminal_cells: u8 (0, 1 or 2)
- populated via unicode-width crate applied to the cluster text
  line[start..end] (harfrust path, uses UnicodeWidthStr) or to the
  single codepoint (no-font fallback path, UnicodeWidthChar)
- layout_to_buffer consumes it when match_mono_width is Some:
      x_advance = cells * mono_width
  instead of the previous round(x_advance / mono_width) * mono_width
  which produced variable cell counts for fallback glyphs.

Covers:
- ASCII + Latin        → width 1  (unchanged visual)
- CJK + fullwidth      → width 2  ✓
- Emoji (incl. ZWJ)    → width 2  ✓  (cluster text handles the ZWJ case)
- Arabic / Hebrew      → width 1  ✓  (was wrongly snapped to 2 before)
- Combining marks      → width 0  ✓  (zero-advance, matches terminals)
- Variation selectors  → width 0  ✓

Limitations: ambiguous-width chars (EAW=A) resolve to 1 via unicode-width
default; a 'cjk' ambiguous mode (unicode-width::UnicodeWidthChar::width_cjk)
could be exposed later as a Buffer flag if needed — not needed for typical
terminal use, matching most wcwidth implementations.

Based on review feedback from lionel@wopr.io on the initial heuristic patch.

Cargo.toml

@@ -30,6 +30,8 @@ sys-locale = { version = "0.3.2", optional = true }
 unicode-linebreak = "0.1.5"
 unicode-script = "0.5.8"
 unicode-segmentation = "1.12.0"
+# Yoda: EAW-aware cell width for monospace/terminal rendering.
+unicode-width = "0.2"
 
 [dependencies.swash]
 version = "0.2.6"