Symmetrically remove prior covariance when initializing from measurement

[ericssonlericson]

Without this change, the covariance matrix generated by SimpleMeasurementInitiatior zeros only rows of the prior covariance. This is a non-symmetric operation, and the resulting covariance matrix thus also becomes non-symmetric. I understand the logic here as "do not add prior covariance to observed dimensions", hence prior covariances between observed dimensions and non-observed should also be removed.

Another option is to only remove prior covariance between observed dimensions, that would be

for dim in mapped_dimensions:
    prior_covar[dim, mapped_dimensions] = prior_covar[mapped_dimensions, dim] = 0