* Interface: -add support for raw parameters (tol,...)
             -custom starting point
             -custom choice of stopping criterion (standard, DIMACS, user-defined,...)
             -add the dummy constraint when A empty

* Problem: -allow the user to save the history of the iterations.

* Workspace: -automate the choice between the additional memory for B (n*m faster but m*m necessary for bigger problems ('eqG32.dat-s')).
              Also investigate numerical errors between the two versions (weird behaviour for 'mcp100.dat-s').
             -Add support for sparse S. Problem: no general Sparse Cholesky is available at the moment. Possible solution:
              compute the sparsity pattern and permutation using SuiteSparse (working only with Float64) and compute the entries
              manually.
              
* IPM: -stopping criterion not sufficient for ill-conditioned problems when tol too small ('control3.dat-s',...) SDPT3 detects it.
        Implement their criterion.
       -switch between cholesky and LU if B numerically not > 0.
       -better exception handling.
       -check infeas at each iteration
       
* Misc: -since 1.7, SparseMatrixCSC constructor explicitly checks for dimensions and allocates. Possible to avoid it?
        -investigate faster alternatives for Extended precision computations (https://github.com/RalphAS/DoubleBLAS.jl,...)
        -add the benchmark results.
        -add documentation
        -encapsulate distinct parts of the code in modules (-> avoid ordering the include's)
        -add more extensive tests
        -add README
       !-'] add MKL' required?
