The subtitles are parsed from each mediaselector playlist in the form of a dict keyed by language each of whose values is a list containing a dict keyed by url and sttl type. There seems to be no attempt in the extractor to make this list contain more than one dict. So a strategy is needed to deal with multiple sttl types for the same language.
For backward compatibility, let the first of any duplicate sttl languages take precedence.