py3: fix mobi ncx

get_id_tag/get_id_tag_by_pos_fid are internal functions that always
return bytes, but when using the former proxied through the latter, we
did not always decode the result in the process of generating an OPF.

As a result, books would end up with nav links pointing to urls
resembling "foo.html#b'anchor'".

Fix by moving down the decode attempt to cover both, right before
writing it back into the index_entries.

src/calibre/ebooks/mobi/reader/mobi8.py

@@ -391,7 +391,7 @@ def create_ncx(self):
                 fi = self.get_file_info(pos)
                 if fi.filename is None:
                     raise ValueError('Index entry has invalid pos: %d'%pos)
-                idtag = self.get_id_tag(pos).decode(self.header.codec)
+                idtag = self.get_id_tag(pos)
                 href = '%s/%s'%(fi.type, fi.filename)
             else:
                 try:
@@ -403,7 +403,7 @@ def create_ncx(self):
                     continue
 
             entry['href'] = href
-            entry['idtag'] = idtag
+            entry['idtag'] = idtag.decode(self.header.codec)
 
         for e in remove:
             index_entries.remove(e)