Lifetime fitter: see if it is possible to re-formulate the math so it can be implemented using symmetric matrices only

The following discussion from !2668 (merged) should be addressed:

• @graven started a discussion:

  it is probably better to not split the problem in a 3+4 dimensional one, but keep it as 7D, at which point the hope is that the relevant matrix is symmetric, and one can just use invChol (and keep it vectorized, and perhaps more numerically stable...) -- but that is for a follow-up MR...