reduce l_icloglock roundtrips

All but one caller of xlog_state_want_sync drop and re-acquire
l_icloglock around the call to it, just so that xlog_state_want_sync can
acquire and drop it.

Move all lock operation out of l_icloglock and assert that the lock is
held when it is called.

Note that it would make sense to extende this scheme to
xlog_state_release_iclog, but the locking in there is more complicated
and we'd like to keep the atomic_dec_and_lock optmization for those
callers not having l_icloglock yet.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <david@fromorbit.com>
Signed-off-by: Niv Sardi <xaiki@sgi.com>