Over the years we have built up a lot of infrastructure to keep delayed
refs in check, mostly by running them at btrfs_end_transaction() time.
We have a lot of different maths we do to figure out how much, if we
should do it inline or async, etc.  This existed because we had no
feedback mechanism to force the flushing of delayed refs when they
became a problem.  However with the enospc flushing infrastructure in
place for flushing delayed refs when they put too much pressure on the
enospc system we have this problem solved.  Rip out all of this code as
it is no longer needed.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>