schur convergence with 0 eigenvalues

[mfalt]

It seems that the _schur! algorithm gets stuck when a row and column is zero

A = [0.0 1.0 0.0;
       -1.0 0.0 0.0;
       0.0 0.0 0.0]
B = [0.0 0.0 0.0;
       0.0 0.0 1.0;
       0.0 -1.0 0.0]

eigvals(BigFloat.(A)) # Max iters reached
eigvals(BigFloat.(B)) # Max iters reached

I fixed this in an incoming PR.
This exposed another bug in _eigvals!(S::Schur{T} where the first "2x2" block in

[0.0 0.0 0.0; 0.0 0.0 1.0; 0.0 -1.0 0]

resulted in 2 zero eigenvalues, which in the end returned all zero eigenvalues.
I fixed this too, however I am not familiar with the algorithm so I might have done something stupid, but it is similar to this fix: #53