fix: handle TOKENIZER_EOF in invalid tag start loop and bare colon check

Two bugs introduced by the NUL truncation fix:

1. Tokenizer_handle_invalid_tag_start: the tag name scanning loop
   checked is_marker() and Py_UNICODE_ISSPACE() to terminate, but
   TOKENIZER_EOF (0x110000) matches neither, causing an infinite loop
   when an incomplete closing tag like "</ref" reaches EOF without ">".

2. Tokenizer_parse colon handling: the bare external link check
   "this == ':' && !is_marker(last)" fired at start-of-input because
   is_marker(TOKENIZER_EOF) returns false (0x110000 not in MARKERS).
   This intercepted ":" before the list handler could run, breaking
   definition list items like ":text" → <dd>.

src/mwparserfromhell/parser/ctokenizer/tok_parse.c

@@ -1863,7 +1863,7 @@ Tokenizer_handle_invalid_tag_start(Tokenizer *self)
     }
     while (1) {
         this = Tokenizer_read(self, pos);
-        if (Py_UNICODE_ISSPACE(this) || is_marker(this)) {
+        if (this == TOKENIZER_EOF || Py_UNICODE_ISSPACE(this) || is_marker(this)) {
             name = Textbuffer_render(buf);
             if (!name) {
                 Textbuffer_dealloc(buf);
@@ -2956,7 +2956,7 @@ Tokenizer_parse(Tokenizer *self, uint64_t context, int push)
             if (Tokenizer_parse_external_link(self, 1)) {
                 return NULL;
             }
-        } else if (this == ':' && !is_marker(last)) {
+        } else if (this == ':' && last != TOKENIZER_EOF && !is_marker(last)) {
             if (Tokenizer_parse_external_link(self, 0)) {
                 return NULL;
             }