Use raw data instead of corrected data for cbf_writer (cctbx#660)

Found by @dermen

For cbf_writer, use imageset.get_raw_data(i) instead of imageset[i], the latter of which applies gain correction.  Solves two problems:

1) Avoids promoting int arrays to floats to save file space
2) Avoids applying the gain twice downstream, since gain is also set on the detector model

Also fix missing encode call for int data

Co-authored-by: Derek Mendez <dermen@lbl.gov>

newsfragments/660.bugfix

@@ -0,0 +1 @@
+Bugfix for dxtbx.image_average: handle detector gain and pixel data type better

src/dxtbx/format/cbf_writer.py

@@ -470,9 +470,9 @@ def add_data_to_cbf(self, cbf, index=None, data=None, verbose=False):
         """
         if data is None:
             if index is None:
-                data = self.imageset[0]
+                data = self.imageset.get_raw_data(0)
             else:
-                data = self.imageset[index]
+                data = self.imageset.get_raw_data(index)
         if not isinstance(data, tuple):
             data = (data,)
 
@@ -539,7 +539,7 @@ def add_data_to_cbf(self, cbf, index=None, data=None, verbose=False):
             if dataisint[i]:
                 elsize = 4
                 elsigned = 1
-
+                byteorder = byteorder.encode()
                 cbf.set_integerarray_wdims_fs(
                     pycbf.CBF_PACKED,
                     binary_id,