incorrect code point output in `decodeHTML`

[6lvkro]

This issue occurs when a complete entity name has no legacy (semicolon-less) form and is followed by a character that cannot continue the name. Since version 7.0.0, it emits a code point unrelated to the name being matched and incorrectly consumes the trailing character.

This behavior was confirmed to work correctly in version 6.0.1.

decodeHTML("&Xi$");   // 6.0.1: "&Xi$"   7.0.0+: "Â"
decodeHTML("&Chi)");  // 6.0.1: "&Chi)"  7.0.0+: "Á"
decodeHTML("&eta8");  // 6.0.1: "&eta8"  7.0.0+: "\uDD05p"
decodeHTML("&Psi-");  // 6.0.1: "&Psi-"  7.0.0+: "pf"

Unlike #2208, which involved a legacy form, this issue has no legacy form.