task 13 : two layers of inexact

[LinaMiao]

No description provided.

[LinaMiao]

-- well, well, well, do not see good result, generally
-- first, reflect mark's newly update in minConf_PQN, generally can not be as good as spgl1 in individual lasso problem.
-- second, inexactness in minConf_spg no theoretical background. Tried two way, optTol in spg varies acording to norm(r), ie, if far away from solution, not solving spg precisely, the other way is like the way testupdate tau, when relative change is within some criteria, accepting the solution, ------ those two method generally do not seem to give robust performance.
-- third, in this code, the stopping criteria of lasso is still not fixed.