[vyncint]

Four emulator-front issues, on top of vt100 0.18.1 (#608).

#580 — text measured in the grid's columns. A new width::columns lays text out exactly as vt100 0.18 does: each character its own width, a zero-width one joining the cell before it, and the presentation selector widening an emoji. Neither unicode-width measure is that — a ZWJ family is one 2-column cluster to UnicodeWidthStr but two wide cells on the grid; ⚠️ is 1 column to a per-character sum but 2 on the grid. A test holds the model against vt100 itself for ten samples (CJK, combining, lam-alef, ZWJ, VS16, keycap, skin tone, flags). Used by:

• insert mode's reservation (the ZWJ family overwrote AB, lam-alef A);
• Screen::locate (reported history column 3 for text at column 5);
• the termlens diff frame padding;
• the text reader, so a saved ⚠️ reads back with the columns it was written at. Releases before this one gave it one column and the text does not say which wrote it: a row that only fits at one column per such cell reads the earlier way, so no file an earlier release wrote stops parsing.

An inserted run that wraps at the margin still overwrites the next row: doing it right needs the scroll region and origin mode, which termlens does not track. Documented in LIMITATIONS.md.

#579 — SGR 58. The grid's normalizer turns 58;5;n / 58;2;r;g;b into one colon parameter, which vt100 skips whole instead of reading 5 as blink and n as an attribute (58;5;0 reset the cell, 58;5;1 made it bold). Still recorded by unsupported(), as ^[[58:5:1m; the allocation-free fast path for plain SGRs is kept.

#581 — CSI ? 7 h is no longer recorded: it asks for autowrap, the emulator's only behaviour. ?7l is still recorded; LIMITATIONS.md says autowrap cannot be turned off.

#585 — documented, as the issue allows: vt100 records a wrap only where the last column held something, and a real fix needs per-row wrap state that survives scrolling, which vt100 has no API for. Screen::row_wrapped and LIMITATIONS.md now say so, with the issue's example.

Verification. Each fix has a test, and reverting it fails that test (insert mode, locate, the pad and the reader each to the string measure; the 58 grouping disabled). Full workspace suite, clippy, rustdoc -D warnings, DCO and attribution gates pass locally. No new PTY-spawning tests.

Closes #579
Closes #580
Closes #581
Closes #585