Remove use of covariance determinant

[eviatarbach]

The covariance determinant is used several times in the package; namely, there is a warning issued if it decreases too much, and it is also used in find_ekp_stepsize to find a stepsize. The first use will be removed by #250, but the second one should also be removed in my view, since the determinant is not a sensible quantity when the covariance is rank deficient (number of parameters > number of ensemble members). Perhaps the trace could be used instead?

[eviatarbach]

In discussion with @odunbar, we decided to remove find_ekp_stepsize altogether, in anticipation of the implementation of timestepping methods in #253.