[Tyler-V]

Fixes #6229

Summary

• Clamp the row walk in BaseSerializeHandler.serialize to terminal.cols
• Clamp StringSerializeHandler._rowEnd to terminal.cols as well: its reads of the row's last two cells and its forced wrap cleanup moves
• Move _terminal into BaseSerializeHandler so the shared row walk can read cols

Problem

A buffer line can be longer than the grid. After a column shrink on the alt buffer, every line keeps its old length (#6228). The serializer walked each row to line.length, so a frame carried cells the grid no longer has and a replay at the current width garbled rows. _rowEnd also read the row's last cells and built the moves that erase its - wrap filler from line.length, so the filler stayed on screen.

Fix

Each row stops at Math.min(line.length, terminal.cols), and _rowEnd uses the same length. serializeAsHTML shares the row walk, so its rows stop at cols too.

Buffer.resize is untouched. #6228 tracks that gap. Clamping in the addon keeps frames on the grid for any path that leaves a line longer than cols.

Test plan

• Unit tests under lines longer than cols: a replay of the SerializeAddon serializes cells past terminal.cols when a line is longer than the grid #6229 repro, and a row ending in empty cells above an autowrapped row, which exercises the separator's cleanup moves
• HTML unit test: a line longer than cols serializes cols cells
• All three fail on master, and the separator test still fails with only the row walk clamped
• addons/addon-serialize/src/SerializeAddon.test.ts: 47 passing
• npm run tsc and npm run lint pass
• The SerializeAddon serializes cells past terminal.cols when a line is longer than the grid #6229 repro against this branch's headless build: replay rows equal source rows

The tests reach a long line through the alt buffer gap in #6228. Once that is fixed they still pass, but they no longer exercise the clamp.

🤖 Generated with Claude Code