id(text/1): TextSurface wrap emits lines wider than the requested width when a single grapheme is wider than the width — contradicting the wrap contract

Evidence

pkgs/id/crates/id-core/src/world_presentation.rs — the new text/1
surface from #2339 (commit 0ec941645):

• wrap_text_line (line ~157) documents: "a word longer than width is
  hard-wrapped on grapheme boundaries so no output line ever exceeds
  width columns (zero-width graphemes attach to the preceding column)".
• The wrap loop only breaks before a grapheme when the current line is
  non-empty (current_width + grapheme_width > width && !current.is_empty()).
  A grapheme whose own display width exceeds width is therefore placed
  whole on its own line, and that line exceeds width.
• TextSurface::with_width documents its result as "already wrapped to
  width columns" and parse_surface accepts a guest width as small
  as 1, so a guest document {"kind":"text","version":1,"width":1, "lines":["界"]} (界 = U+754C, unicode-width 2) validates and produces
  a stored line of width 2 at width 1.

Verified this run with a laptop probe at origin/main c19c1d8b0
(worktree probe-2026-11-03-laneB-scan19, logs /tmp/probe-scan19.log
and /tmp/probe-scan19b.log on the laptop, read back from disk):

• TextSurface::with_width(1, vec!["界"]) → Ok, lines == ["界"]
  (display width 2 > requested width 1)
• parse_surface("{\"kind\":\"text\",\"version\":1,\"width\":1, \"lines\":[\"界\"]}") → Ok(Some(Text(_))) with the same over-wide line
• Control: with_width(2, vec!["a界b"]) wraps to lines within width 2,
  and the existing tests at width 3 (ab界cd) stay within width — the
  violation needs a grapheme strictly wider than the wrap width itself.

Why it matters

text/1 is the main presentation mode for agents and scripts (#2339):
presenters size their output to the negotiated width and trust the
validated surface to fit it. A validated line that exceeds the width
breaks that contract exactly where it is hardest to see — narrow
terminals / width-1..2 viewports and wide CJK/emoji graphemes — and the
module's own doc states the opposite guarantee, so downstream code has
no reason to re-check. The screen side never faces this because a
ScreenSurface cell names its column span explicitly; the text side
needs its own rule.

Acceptance criteria

• Code and docs agree, one of:
  (a) TextSurface::with_width / parse_surface refuse a line whose
  single grapheme is wider than the requested width (naming the
  width in the error), or
  (b) the wrap contract is amended in the module docs to state that a
  line may exceed width by the width of one unsplittable grapheme,
  and presenters are pointed at that exception.
• Regression tests cover: width 1 with a width-2 grapheme, width 2 with
  a width-2 grapheme (accepted, fits exactly), and the existing
  width-3 ab界cd case still passing.
• cargo test -p id-core --lib world_presentation passes (laptop or CI).

Relates to #2339 (the text/1 contract issue) — this is a specific
wrap-contract violation in the landed implementation, not covered by
that issue's remaining work.