Fix handling of "ambiguous ampersands" to report error when expected

Fixes #82

44427c0 broke handling of "ambiguous
ampersands" for the case where we've already examined enough of string
after the ampersand to conclude that the substring we have so far
doesn't match the beginning of any known named character reference.

This change re-conforms the error-reporting for that case to the
requirements in the HTML spec. Otherwise, without this change, no error
is reported as expected in many or most cases where an ampersand doesn't
actually start a character reference.

src/nu/validator/htmlparser/impl/Tokenizer.java

@@ -3529,9 +3529,13 @@ private void ensureBufferSpace(int inputLength) throws SAXException {
 
                     if (candidate == -1) {
                         // reconsume deals with CR, LF or nul
-                        if (c == ';') {
-                            errNoNamedCharacterMatch();
-                        }
+                        // We can go ahead and emit an error here without
+                        // needing to do an "if (c == ';')" check, because if
+                        // we've gotten here, we've already examined enough of
+                        // string after the ampersand to determine that the
+                        // substring we've looked at so far doesn't match the
+                        // beginning of any known named character references.
+                        errNoNamedCharacterMatch();
                         emitOrAppendCharRefBuf(returnState);
                         if ((returnState & DATA_AND_RCDATA_MASK) == 0) {
                             cstart = pos;