Upgrade to Tika 1.21

   * Add optional AUTO mode to OCR'ing of PDFs.  If tesseract is installed
     and on the path, and this option is selected programmatically
     or via TikaConfig(), the PDFParser will use heuristics to decide
     whether or not to run OCR per page on PDFs. (TIKA-2749)

   * The ZipContainerDetector's default behavior was changed to run
     streaming detection up to its markLimit.  Users can get the
     legacy behavior (spool-to-file/rely-on-underlying-file-in-TikaInputStream)
     by setting markLimit=-1. The POIFSContainerDetector requires an underlying file;
     it will try to spool the file to disk; if the file's length is > markLimit,
     it will not attempt detection; set markLimit to -1 for legacy behavior (TIKA-2849).

   * Upgrade PDFBox to 2.0.14 (TIKA-2834).

   * Add CSV detection and replace TXTParser with TextAndCSVParser;
     users can turn off CSV detection by excluding the TextAndCSVParser
     and adding back the TXTParser via tika-config (TIKA-2833).

   * Add a CSVParser.  CSV detection is currently based solely on filename
     and/or information conveyed via Metadata (TIKA-2826).

   * General upgrades: asm, bouncycastle, commons-codec, commons-lang3, cxf,
     guava, h2, httpcomponents, jackcess, junrar, Lucene, mime4j, opennlp, parso,
     sqlite-jdbc (provided), zstd-jni (provided) (TIKA-2824)

   * Bundle xerces2 with tika-parsers (TIKA-2802).

   * Upgrade jaxb to 2.3.2 (TIKA-2819).

   * Upgrade jackson to 2.9.8 (TIKA-2717).

   * Update tika-eval's common tokens lists (TIKA-2822).

   * Handle bad tags in tika-eval more robustly (TIKA-2810).

   * Add reports for tags in tika-eval (TIKA-2809).

   * Extract text from SDT element within textboxes in .docx files (TIKA-2807).

   * Try to handle truncated OOXML files more robustly (TIKA-2765).