Fix a tokenizer bug causing incorrect handling of duplicate attributes

The tag buffer was retaining duplicate attribute values that were marked
to be discarded, thus clobbering the following attribute name.

This also fixes the relevant test, which didn't catch this particular
case because it had an empty (boolean-style) value as the duplicate.

src/tokenizer.c

@@ -815,6 +815,7 @@ static void finish_attribute_value(GumboParser* parser) {
     // Duplicate attribute name detected in an earlier state, so we have to
     // ignore the value.
     tag_state->_drop_next_attr_value = false;
+    reinitialize_tag_buffer(parser);
     return;
   }
 

tests/parser.cc

@@ -285,7 +285,7 @@ TEST_F(GumboParserTest, Whitespace) {
 }
 
 TEST_F(GumboParserTest, DuplicateAttributes) {
-  std::string text("<input checked=\"false\" checked id=foo id='bar'>");
+  std::string text("<input checked=\"false\" checked=true id=foo id='bar'>");
   Parse(text);
 
   GumboNode* body;