Use normwise criterion for inf eigenvalues in QZ

[thijssteel]

trafficstars

As described in "A new deflation criterion for the QZ algorithm" T. Steel, R. Vandebril and J. Langou (https://arxiv.org/abs/2208.02057) When testing for infinite eigenvalues in the QZ algorithm, the normwise criterion performs better than the elementwise one.

This PR changes the infinite eigenvalue check to the normwise criterion, but keeps the elementwise check for the finite eigenvalues.

[angsch]

Looks good to me :-)

With your changes tempr becomes an unused variable in [cz]laqz0, if you want to remove it.