Snapshot box misaligns for rows holding emoji sequences

Nobody has claimed this yet.

A snapshot frame is drawn to the grid's column count, but the grid measures a character at a time while a terminal renders an emoji sequence as one glyph. A row holding one comes out narrower than its frame.

Reproducing, with no window title involved:

```
Str-width= 14 ╭────────────╮
Str-width= 8 │👨‍👩‍👧‍👦ab │ ← six columns short
Str-width= 14 │ │
Str-width= 14 ╰────────────╯
```

The grid puts that family emoji in 8 columns (one cell plus 4 continuations) because `alacritty_terminal` measures with `UnicodeWidthChar` per character. A terminal draws it in 2, which is what `UnicodeWidthStr` reports for the whole string. The same gap affects skin-tone modifiers, keycaps, and a base character followed by a variation selector.

Measured on `unicode-width` 0.2.2:

| sequence | per character | whole string |
| --- | --- | --- |
| `👨‍👩‍👧‍👦` | 8 | 2 |
| `👍🏽` | 4 | 2 |
| `1️⃣` | 1 | 2 |
| `❤️` | 1 | 2 |

This predates #91 and is independent of it: #91 only made the *title* measure whole strings, since it is a string we place rather than grid content. Fixing the rows means changing how the grid assigns cells to a sequence, which is a decision about the emulator seam rather than about the serializer, and would want the same treatment in every backend.

Worth deciding whether shell-use follows its emulator here or groups sequences itself before the two disagree somewhere more visible than a snapshot border.

No file or test path is named. Start at the grid’s per-character cell assignment and the alacritty_terminal UnicodeWidthChar measurement, then reproduce the family-emoji snapshot case. Done requires an agreed treatment for emoji sequences that keeps snapshot borders aligned consistently across backends.

Written by the indexing model from the issue text.