Fix cram_dependent_data_series when FN is used.

If we get into the primary loop of cram_decode_seq, iterating over
feature (count FN) then we have unguarded code that always checks for
sequence overlapping the reference.  To do this, seq_pos must be set.
Some data series we know how seq_pos is adjusted irrespective of
whether we decode, eg BS is always +1, but others are strings and the
only way we know how to update seq_pos is to decode them.

Hence added FN as having a dependency on SC, IN and BB.

Fixes samtools/samtools#1475

cram/cram_decode.c

@@ -668,6 +668,13 @@ int cram_dependent_data_series(cram_fd *fd,
             s->data_series |= CRAM_CF | CRAM_NF;
         if (s->data_series & (CRAM_BA | CRAM_QS | CRAM_BB | CRAM_QQ))
             s->data_series |= CRAM_BF | CRAM_CF | CRAM_RL;
+        if (s->data_series & CRAM_FN) {
+            // The CRAM_FN loop checks for reference length boundaries,
+            // which needs a working seq_pos.  Some fields are fixed size
+            // irrespective of if we decode (BS), both others need to know
+            // the size of the string fetched back (SC, IN, BB).
+            s->data_series |= CRAM_SC | CRAM_IN | CRAM_BB;
+        }
 
         orig_ds = s->data_series;