Try to improve the performance of thin_ls by incorporate the new io method used by thin_check. Expected performance improvement is about 20% to 30%.