Solve rare FPEs in MET trigger code

Extremely rarely it's possible to have a singular pileup covariance matrix in the pufit algorithm. This causes FPEs when we try to invert it, so this MR introduces a check for this and then returns the expected corrections from before the fit is run. While looking into this I noticed some other UB in our code where we use std::atan2 - the result of this isn't defined if both inputs are 0 so I also fixed this.

Solves ATR-24694