description box: right border is offset for CJK / wide characters (padded by code units, not display width)

Describe the bug

The suggestion description box (the right-hand box) is padded with String.prototype.padEnd, which counts UTF-16 code units, while renderBox() draws the border assuming the row occupies exactly width terminal columns. Whenever a description contains characters whose display width differs from their code-unit length — CJK text, BMP-wide emoji (e.g. U+2B50), ZWJ emoji sequences — the right border is offset: pushed out for wide characters, pulled in for ZWJ sequences.

The suggestion name box is unaffected, because truncateText() already pads with wcPadEnd()/wcswidth().

To Reproduce

1. Add a local spec via specs.path:

~/.config/inshellisense/rc.toml

[specs]
path = ["/Users/<you>/.config/inshellisense/specs"]

~/.config/inshellisense/specs/index.js

const specs = ["opencode"];
const diffVersionedCompletions = [];
export { specs as default, diffVersionedCompletions };

~/.config/inshellisense/specs/opencode.js

export default {
  name: "opencode",
  description: "OpenCode command line interface",
  subcommands: [
    { name: ["upgrade", "update"], description: "⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐" },
    { name: "auth", description: "以一条消息运行 OpenCode" },
  ],
};

2. Start is and type opencode up

Actual behaviour

The description box's right border does not line up with the top border. With 10 × U+2B50 the row is padded to 38 columns instead of 28:

Expected behaviour

Every row is padded to width display columns, so the box borders stay aligned, exactly like the name box.

Root cause

src/ui/utils.ts:37

return truncatedLines.map((line) => line.padEnd(width));

padEnd pads to width code units, but renderBox() (line 15) emits "│" + row + "│" assuming row is width columns wide. Measured with the repo's own wcswidth() (width = 28, i.e. descriptionWidth - borderWidth):

sample	code units	display cells	padEnd → cells	wcPadEnd → cells
Run OpenCode with a message	27	27	28 ✓	28 ✓
以一条消息运行 OpenCode	16	23	35 ✗	28 ✓
⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐ (U+2B50)	10	20	38 ✗	28 ✓
🚀🚀🚀🚀🚀🚀🚀🚀 (U+1F680)	16	16	28 ✓ (coincidence)	28 ✓
👨‍👩‍👧👨‍👩‍👧👨‍👩‍👧	24	18	22 ✗	28 ✓

Note that astral emoji happen to match (2 code units = 2 columns), which is why the bug can look like it only affects CJK; ZWJ sequences fail in the opposite direction.

Suggested fix

Use the display-width helper that already exists in the same file and is already used by truncateText():

return truncatedLines.map((line) => wcPadEnd(line, width));

ASCII descriptions are unaffected (wcswidth(text) === text.length for them). PR to follow.

Environment

• inshellisense 0.0.4 (Homebrew), is --version → 0.0.4
• macOS 27.0 (26A428)
• zsh
• VS Code integrated terminal

Related: #175, #302

Start in src/ui/utils.ts at line 37 and read renderBox() near line 15, then compare the description padding with truncateText() and the existing wcswidth() helper. Reproduce with the CJK, wide-star, and ZWJ descriptions from the issue. Done means every description row occupies the configured display width and the right border aligns with the top border.