The recover flag to LibXML, if you really must, or XML-Liberal if you really want to go overboard in parsing any old rubbish.
I'm sure you would like to ignore the question of whether parsing non-well-formed documents makes any sense, but ignoring it won't make it go away. Most RSS tools will correctly reject any non-well-formed XML input completely; you should generally follow suit, unless your tool is something unusual like an RSS debugger.
“Tag soup” is a term specifically related to HTML parsing. One of the central ideas of XML (and hence RSS and Atom) is that there is no such thing.
