iterative ref: learn white noise (sigma) iteratively

Should be able to implement inside compute_bayesian_weights

after compute bayes weights
weight residual norm.
- have part of residual from log_bayesian_weights = scale * (slices_norm - 2 * corr_slices_particle)
- just need to add in norm of X (data). can precompute
- divide through by all particles and pixels, and update at end of round. so can compute sigma_i and divide by the number of particles at the end

compSPI / reconstructSPI