- somehow it happens that spreadsheets sometimes think they have data
  for 100's or 1000's of rows/columns even though all cells are empty.
- xls / xlsx processing refactored to stop processing after 20 adjacent
  empty columns (when getting the headers) or rows (when reading rows).
- add tests for xls / xlsx to show bad input processed as expected.
- the change to not output empty rows seems to have broken the
  "flatxlsformtest" case which needs further investigation - seems to
  just be a question numbering issue and different itext lang order.