fix: align JSX expression line handling with upstream

[fansenze]

Summary

Fix react/jsx-one-expression-per-line so literal and single-child allowances work when JSX follows a line break or comment. For example, this remains valid with allow: "literal":

const view = () => (
  <A>text</A>
);

Preserve complete trailing ASCII-space runs, distinguish LF from bare CR when fixing multiline text, and exclude disabled reports from adjacent-fix selection. Reuse tsgo's trivia scanner and rslint's ECMAScript whitespace helpers; avoid reversing entire text nodes, sorting children already in source order, and computing fix selection for diagnostics-only requests.

Validation:

• Compared 10,097 upstream and generated cases against the latest published eslint-plugin-react@7.37.5 using ESLint 8.57.1 and 10.10.0: zero differences in diagnostic messages, positions, counts, or final multipass output. The ESLint 10 driver delegates its removed SourceCode.isSpaceBetweenTokens method to the unchanged ESLint 8 implementation; unused-disable reporting is disabled to isolate this rule.
• go test ./internal/plugins/react/rules/jsx_one_expression_per_line -count=3 passed, including regression cases and edit-demand checks.
• The corresponding JavaScript rule suite and eight real CLI autofix cases passed, including BOM preservation, CR, and disable directives.
• pnpm run check-spell, pnpm run format:check, and pnpm exec golangci-lint run --new-from-merge-base=origin/main ./internal/plugins/react/rules/jsx_one_expression_per_line/... passed. Go lint was limited to this package and introduced code.

Go benchmark medians from five 500 ms samples on Apple M5 Max, Go 1.26.5, with -benchmem -cpu=1. Parsing and JSX-node collection were outside the timed loop; the benchmark exercised rule setup and listeners with different edit demands. The temporary benchmark source was removed.

Case	ns/op before → after	B/op before → after	allocs/op before → after
NoJSX	175.8 → 174.3	512 → 512	5 → 5
Formatted	10,586 → 10,176	11,104 → 11,080	112 → 111
Diagnostics	37,782 → 30,686	75,232 → 70,272	574 → 569
Autofix	46,124 → 40,422	82,400 → 82,392	958 → 957
LongText	73,795 → 292.4	156,192 → 520	8 → 6
NonJSXAllowed	338.9 → 345.9	512 → 512	5 → 5
SingleChildAllowed	191.4 → 198	512 → 512	5 → 5

Diagnostics and autofix improved by 18.8% and 12.4% respectively. The long-text case avoids a full-text allocation and scan; lightweight control cases varied by less than 4%, so no end-to-end performance claim is made.

Related Links

• Upstream rule documentation
• Upstream implementation

Checklist

• Tests updated (or not required).
• Documentation updated (or not required).