Various QR decoder fixes to bring it more in-line with the spec.

- Support mirror-image reversals.
  We still don't support reflectance-reversals.
- Properly compute the self-parity of each symbol in an S-A group.
  We still don't use it to validate the parity of the entire group.
- Properly decode the Application Indicator for FNC1 in the 2nd
   position, and prepend its textual representation to the output.
- Only use the first S-A header, instead of the last, as this is
   supposed to be the very first entry in the stream.
- Only use the first FNC1 marker, instead of the last, as this is
   supposed to appear once, before any data entries (after S-A and
   ECI).
- Properly reserve some of the parity bytes for small codes
   (versions 1, 2-L, and 3-L) for error detection instead of
   correction, as the spec requires.
  This causes zxing-4/43.png to fail where it succeeded before.
- Add support for the UTF-8 ECI marker (000026).
- Treat data encoded in kanji mode as part of the ECI-specified
   character set, if any, instead of always treating it as SJIS.

Also, assorted bug fixes:
- Properly reset the "empty" count after finding an isolated module
   while scanning the right edge during homography estimation.
- Avoid division by zero when projecting the lower-right alignment
   pattern out to the corner during homography estimation.
- Limit the number of consecutive invalid configurations we'll try,
   to avoid long delays in images with lots of finder-pattern-like
   features.
- Properly report the FNC1 markers back to zbar via the modifiers
   field.

zbar/decoder.h

@@ -28,6 +28,8 @@
 
 #include <zbar.h>
 
+#include "debug.h"
+
 #define NUM_CFGS (ZBAR_CFG_MAX_LEN - ZBAR_CFG_MIN_LEN + 1)
 
 #ifdef ENABLE_EAN

zbar/decoder/qr_finder.c

@@ -51,6 +51,10 @@ zbar_symbol_type_t _zbar_find_qr (zbar_decoder_t *dcode)
     qrf->s5 += get_width(dcode, 1);
     s = qrf->s5;
 
+    /*TODO: The 2005 standard allows reflectance-reversed codes (light on dark
+       instead of dark on light).
+      If we find finder patterns with the opposite polarity, we should invert
+       the final binarized image and use them to search for QR codes in that.*/
     if(get_color(dcode) != ZBAR_SPACE || s < 7)
         return(0);