Multiple fixes to allow PDFio to read more edge-case PDFs.

- Update _pdfioFileGets to allow for really long lines where it
  doesn't matter if we lose the end of the line.
- Update "startxref" detection at the end of the file.
- Refactor repair logic so that you just get a single WARNING about
  the repair (debug messages available for testing)
- Allow whitespace after the "obj" in the object header.
- Make sure to close xref stream on error.
- Update predictor code to support Colors <= 32 (some implementations
  set Colors to the number of bytes per record in the xref stream,
  which prevents the predictor from doing anything...)
- Allow CR CR in xref table.
- Clear old trailer/root/pages/etc. objects when repairing, update
  existing objects that were already found in load_xref.
- Don't set current object in pdfioObjectCreate/OpenStream if the
  stream can't be created/opened.

test-corpus.sh

@@ -18,12 +18,22 @@ if test $# = 0; then
 fi
 
 for file in $(find "$@" -name \*.pdf -print); do
+    # Don't worry about test files containing MIME garbage...
+    (head -4 $file | grep -q Content-Type) && continue;
+
+    # Or test files containing MacBinary garbage...
+    (file $file | grep -q MacBinary) && continue;
+
+    # Don't worry about test files that Xpdf can't handle...
     pdfinfo $file >/dev/null 2>&1 || continue;
 
+    # Run testpdfio to test loading the file...
     ./testpdfio $file >$file.log 2>&1
     if test $? = 0; then
+    	# Passed
         rm -f $file.log
     else
+    	# Failed, preserve log and write filename to stdout...
         echo $file
     fi
 done