Extend the LRS to serve as a conversion device

[claus-zinn]

The Apache Tika language guesser is unreliable for PDF, RTF and DOCX documents. Therefore, the LRS first converts these documents to plain text (also using Apache Tika) before the language guesser is started. However, the LRS matcher uses the original mime type of the document to identify applicable tools (and the original document is sent to the selected tool). As a consequence, fewer tools are found applicable. In the future, allow users to decide whether they want the tools to work on the original mime types, or on text/plain documents (hence increasing the set of applicable tools).