fix: handle left-angle-bracket that's not a tag (#733)

This handles the case where there's a `<` followed by one or more
character such that it looks like the beginning of a start tag and then
a space and then a thing that actually is a start tag. Something like:

```
<tag <b>text</b>
```

This fixes it by identifying the situation and then shoving everything
after the space back into the character stream to get parsed again.

Author: willkg

bleach/html5lib_shim.py

@@ -322,6 +322,7 @@ def __iter__(self):
         last_error_token = None
 
         for token in super().__iter__():
+            print(token, last_error_token)
             if last_error_token is not None:
                 if (
                     last_error_token["data"] == "invalid-character-in-attribute-name"
@@ -346,6 +347,36 @@ def __iter__(self):
                     last_error_token = None
                     yield token
 
+                elif (
+                    last_error_token["data"] in (
+                        "invalid-character-in-attribute-name",
+                        "invalid-character-after-attribute-name",
+                    )
+                    and token["type"] == TAG_TOKEN_TYPE_CHARACTERS
+                    and token.get("data")
+                    and " " in token["data"]
+                ):
+                    # token["data"] has something that starts with a left angle
+                    # bracket, then has some characters followed by a space
+                    # followed by another left angle bracket and ending with
+                    # a right angle bracket. That part could be a real tag, so
+                    # we don't want it to get treated as Characters. For
+                    # example, soemthing in this shape: <nottag <...>
+                    # If so, we want to take off the first bit that is
+                    # definitely not a tag and reparse the rest.
+                    head, rest = token["data"].split(" ", 1)
+                    if rest.strip().startswith("<"):
+                        # yield the not-a-tag plus the space we split on
+                        token["data"] = head + " "
+                        yield token
+
+                        # shove the rest back in the stream for the praser to look
+                        # at
+                        for c in reversed(rest):
+                            self.stream.unget(c)
+                    else:
+                        yield token
+
                 elif (
                     last_error_token["data"] == "expected-closing-tag-but-got-char"
                     and self.parser.tags is not None