How it worked before:

  * Extract the content without allowing any `}` in it, and require that it ends with `}` - for block elements anchored to the end of the line, otherwise not.
  * Parse the content in more detail. No edge cases with `}` can arise. If parsing is interrupted by some unrecognized token, discard the rest of the string.

How it works now:

  * Extract the content *and allow* `}` in it, and require that it ends with `}` - for block elements it's anchored to the end of the line, otherwise not.
  * Parse the content in more detail. Allow `}` only within the quoted parts, otherwise interrupt parsing like for any other unrecognized token.
    If parsing is interrupted, there is remaining unrecognized text. Ideally perhaps we would bail out at this point entirely (and not recognize it as an attr_list), but to preserve historic behavior, any extra text before `}` is just discarded.
    If there is an extra `}` in the remaining text:
      * For block elements: that must mean that the attr_list syntax did not in fact terminate at the end of the line but earlier. So, bail out and do not register any attributes and do not change the original text.
      * For inline elements: that must mean that we just overmatched a bit, but that's OK, we just assign attrs as normal and put the extra text back into the string. As mentioned, any extra text *before* `}` is just discarded.