fix(text): truncate by grapheme cluster, like DisplayWidth measures

[yu2971512385-ui]

Truncate decides whether a string needs cutting with DisplayWidth
(lipgloss.Width, which measures grapheme clusters), then performs the cut with
reflow/truncate, which measures rune by rune. The two disagree about anything
assembled from more than one rune, so the result misses the budget it was given.

The symptom

A table at a 46-column terminal width, rendered with pkg/tableprinter:

# trunk
#1201  fix: handle 👨‍👩‍👧‍👦 family na...        open
#1202  docs: add 🇯🇵 Japanese translat...  open
#1203  chore: bump dependencies after...  open

# with this change
#1201  fix: handle 👨‍👩‍👧‍👦 family names in...  open
#1202  docs: add 🇯🇵 Japanese translat...  open
#1203  chore: bump dependencies after...  open

A terminal draws 👨‍👩‍👧‍👦 in two columns, and lipgloss.Width agrees.
A rune-wise walk sees four wide emoji joined by zero-width joiners and charges
eight. So Truncate hands back six fewer columns than it was asked for, and the
open column on that row lands six places right of where it does on every other
row.

The error runs the other way for regional indicator pairs, which are two runes
of width 1 each but one two-column glyph:

text.Truncate(1, "🇯🇵")  // "🇯" — half a flag, two columns wide, budget was 1

The change

ansi.Truncate comes from charmbracelet/x/ansi, which is what lipgloss.Width
already calls to produce the number DisplayWidth returns, so the measurement
and the cut can no longer drift apart. It was already in the module graph as an
indirect dependency, and it handles the ANSI-escape preservation reflow was
doing (reflow stays in the graph via glamour).

This finishes what #159 started: that PR moved DisplayWidth onto lipgloss to
fix cli/cli#9018, but left the cut itself on the old width model.

Tests

Every existing case in TestTruncate passes unchanged — including the Japanese,
Chinese, Korean, accented-character and ANSI-coloured ones — plus two new cases
for a ZWJ sequence and a regional indicator pair.

TestTruncateFillsTheBudget is the invariant underneath all of them: across a
corpus of ASCII, CJK, emoji, combining-mark and coloured strings at every width
from 1 to 24, DisplayWidth(Truncate(n, s)) is never above n, and is exactly
n whenever the input was genuinely too wide. It records 97 assertion failures
on trunk and none after the change.

🤖 Generated with Claude Code