[tlysanhuo]

Recording a contenteditable editor used textContent, so typing Alpha, Enter, Beta produced a final Trace v3 fill of AlphaBeta. Preserve Alpha\nBeta and keep blank/trailing lines without treating a caret placeholder <br> as extra content.

Fixes #594.

Extract the existing fill verification reader into a self-contained shared function and use it for recording as well. Native inputs and textareas retain their .value; simple editable lines retain their existing whitespace/placeholder handling, and rich structures retain rendered-text handling. The fill tool serializes the same function into the page. Includes English and Chinese changelog entries.

Validation on Node.js 22.23.4 / Chrome for Testing 157.0.8103.38 (Linux):

• The 14 new regression cases against unchanged main: 10 failed, 4 passed. With the fix, those cases plus the existing fill browser suite: 182 passed.
• Real browser cases cover Enter, Shift+Enter, blank lines, trailing newline, rich paragraphs, pasted newlines, plaintext-only editing, textarea values, redaction, empty placeholders and clearing an editor. The minified reader is also serialized back into the page and checked.
• Full default extension suite: 3,699 passed, 428 opt-in tests skipped. The new browser suite and existing fill browser suite were run explicitly as described above.
• Extension TypeScript check, production build and pnpm lint passed. The lint command includes DSH checks: 713 passed, 10 skipped; dependency source-map warnings were non-fatal.
• Changelog synchronization and git diff --check passed.

Run the browser regressions with:

BSK_CLICK_CHROME=/path/to/chrome BSK_FILL_CHROME=/path/to/chrome pnpm --filter @browser-skill/extension exec vitest run src/content/__tests__/record-contenteditable.browser.test.ts src/tools/__tests__/fill.browser.test.ts

The recording browser tests bundle the production capture, step buffer and Trace v3 reducer, using native CDP input in an isolated browser/profile. They supply synthetic observation state IDs; they do not cover the installed extension's transport, observation matching, daemon or CLI. Rust tests were not run because no Rust code changed. Browser regressions are opt-in; this PR does not change CI workflows.

Implemented and tested with Codex assistance; no independent human review is claimed.