Fix JS decoding of multiline sequences

The previous logic updated the selection's end position at every pass,
which tended to miss codes at line ends (especially if a final
newline was absent); e.g.

    \u274C\u2705\u274C\u2705\u274C\u2705\u274C\u2705\u274C\u2705
    \u274C\u2705\u274C\u2705
    \u274C\u2705
    \u274C

decoded as

    ❌✅❌✅❌✅❌✅❌✅
    ❌✅❌✅
    ❌\u2705
    \u274C

An integer overflow exception when encoding 4-byte glyphs like 🍪 was
quietly fixed in ad2e69c. More work still needed before utf8mb4 can be
encoded *correctly*

Author: Robert Di Pardo

src/U_JSEncode.pas

@@ -119,7 +119,6 @@ function DecodeJS(Scope: TEntityReplacementScope = ersSelection): Integer;
       if Assigned(Match) then begin
         // Adjust the target already
         Target.StartPos := Match.StartPos + 1;
-        Target.EndPos := Target.EndPos - Length(Match.Text) + 2;
 
         // replace this match's text by the appropriate Unicode character
         Match.Text := Char(StrToInt('$' + Copy(Match.Text, 3, 4)));