Fix: archfiends generated in bones weren't tracked properly

From a github report, Geryon was loaded in a bones file, and
subsequently killing him did not get rid of his debuff during the
ascension run. The underlying cause is that the num_in_dgn counter did
not tick up when an archfiend was loaded from a bones file, which
subsequently caused the counter to underflow to UINT_MAX when they were
killed. Since that number is greater than 0, fiend_adversity concluded
they were still active in the dungeon.

Fix that problem, and also add a bounds check to guard against the
underflow by printing an impossible if for some reason num_in_dgn
attempts to decrement when already at 0.

Fixes #213

src/mon.c

@@ -2867,7 +2867,12 @@ m_detach(
     }
     if (is_archfiend(mtmp->data)) {
         struct fiend_info *fiend = lookup_fiend(monsndx(mtmp->data));
-        fiend->num_in_dgn--;
+        if (fiend->num_in_dgn == 0) {
+            impossible("already 0 of this archfiend?");
+        }
+        else {
+            fiend->num_in_dgn--;
+        }
     }
 
     if (mtmp->m_id == gs.stealmid)

src/restore.c

@@ -1160,6 +1160,12 @@ getlev(NHFILE *nhfp, int pid, xint8 lev)
                                        == sgn(mtmp->data->maligntyp))) ? 1
                                   : peace_minded(mtmp->data);
             set_malign(mtmp);
+            /* archfiends that newly exist in the game due to this bones file
+             * need to be counted */
+            if (is_archfiend(mtmp->data)) {
+                struct fiend_info *fnd = lookup_fiend(monsndx(mtmp->data));
+                fnd->num_in_dgn++;
+            }
         } else if (elapsed > 0L) {
             mon_catchup_elapsed_time(mtmp, elapsed);
         }