Never split an emoji at the edge of a find snippet or a cut read

[kevin9327]

#115 stopped the compact view's cap from printing half of a character past U+FFFF. Two other cuts in src/act.js count UTF-16 units the same way and still split one:

• find's snippet window (search) opens 60 units before the match and runs 200.
• read's cut of a first block bigger than its whole budget stops at budget * 4 units.

Either edge can land between the two halves of an emoji, and the line then carries a lone surrogate, which a terminal prints as �. A block with the match 60 units after an emoji, and another emoji 200 units on, printed as oc find needle on a page with enough matches to use snippets:

main:   "[1] ... \ude00xxx...needleyyy...\ud83d ..."   isWellFormed: false
branch: "[1] ... 😀xxx...needleyyy... ..."             isWellFormed: true

Change. Each edge steps back one unit when it would land between the halves, the rule render.js already uses at the cap. The snippet's opening edge so takes the whole emoji, and its closing edge and read's cut leave it out. The window width, the ... markers, and the read budget are otherwise unchanged.

Test. tests/act.test.js gains one test covering both: a snippet whose window opens and closes inside an emoji, and a read --budget 10 whose cut at character 40 lands inside one. On main the snippet half fails; with only the snippet fixed, the read half still fails (xxx\ud83d ... cut at ~10 tokens), so each half guards its own edge. Full suite on Windows (Node 24): 316 tests, 313 pass, 3 skipped, 0 fail.

Left alone. The JSON view's same on every item line also cuts a value at 60 units (v.slice(0, 60) in jsonToHTML). It is a different view, so I kept it out of this change. No CHANGELOG line, to keep this round's PRs from conflicting there.

🤖 Generated with Claude Code