avcodec/jpeg2000dec: Use 32x32->64bit for 5/3 dequantization

This fixes overflows, using fewer bits would impact high bit depth quality

Fixes Ticket4654
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

libavcodec/jpeg2000dec.c

@@ -1437,7 +1437,7 @@ static void dequantization_int(int x, int y, Jpeg2000Cblk *cblk,
         int32_t *datap = &comp->i_data[(comp->coord[0][1] - comp->coord[0][0]) * (y + j) + x];
         int *src = t1->data[j];
         for (i = 0; i < w; ++i)
-            datap[i] = (src[i] * band->i_stepsize) / 32768;
+            datap[i] = (src[i] * (int64_t)band->i_stepsize) / 32768;
     }
 }