[johnhldiver-beep]

Fixes #353.

Parsing <p>a &amp; b</p> produced 3 Text nodes ("a ", "&", " b") instead of 1 ("a & b"), because htmlparser2 with decodeEntities: true calls ontext once per piece around each character reference and the handler appended a new Text node per call. The HTML parsing algorithm appends each character to the previous Text node ("insert a character"), and jsdom returns one node here.

Fix (in esm/shared/parse-from-string.js plus the identical logic in the committed cjs/ mirror): merge an ontext piece into the previous sibling when it is a Text node (prev[VALUE] += text), otherwise appending a new node as before. CDATA handling is untouched; nothing about serialization, escaping, or normalize() changed.

Tests: regression tests in test/html/document.js covering entity merge, multi-entity merge, plain text unchanged, and no merging across element or comment boundaries. New tests failed before the fix ("Expected: 1, Got instead: 3") and pass after; test/html/document.js and the full node test/index.js suite both exit 0 with no failures. Honest limits: this suite prints no numeric assertion totals, so counts are reported as exit codes + failure-grep counts; eslint/tsc/rollup were not run (devDeps deliberately skipped for disk).

Note: the issue reporter offered in the issue body ("If yes, I can open a PR") but has filed nothing as of this PR — no PR by number or concept covers this fix.