Measurement observation may be zero when no measurement available

[ChristopherRabotin]

Bug report

Describe the bug

If a ground station is configured to support N measurement types but only M<N are requested, then one of the entries of the observation vector is zero, as expected. However, if the filter is set to process measurements sequentially, then it may try to process zero as a real observation, thereby causing a prefit of exactly the opposite of the expected/computed measurement.

#409 introduced a hack fix to prevent this, but a better fix is warranted.

Similarly, it would be useful for the Residual type to also include the real and computed observations so that all runs are effectively a "residuals versus reference" run as well.

Expected behavior

The zeroed observation should be skipped (and they are with a hack).