Warnings for Velo backward tracks fitted with PrKalmanFilter

Although the central part of Alignment around AlignAlgorithm was already adapted to PrKalmanFilter, there are still some unexplained warnings when using Velo backward tracks fitted with PrKalman to align Velo. An example of such a warning is

HumboldtAlignAlgorithm              WARNING found bad element on offdiagonal: 4 0 -0.185941 -1.00166
HumboldtAlignAlgorithm              WARNING Track type = 'Velo' 'PrPixel' 'Backward'
HumboldtAlignAlgorithm              WARNING Error computing residual cov matrix. Skipping track of type 'Velo' with key: 19 and chi2 / dof: 0.000173338/1

The warning about a bad element on offdiagonal is invoked here in AlResiduals.h, the problem seems to be related to V - HCH somehow. The other warnings are just consequences of the first one.

The hope was that fixing the fit of Velo backward tracks with PrKalmanFilter in Rec!2734 (merged) would help (reversing the order of nodes, fixing downstream seed state, and adding the backward flag to backward tracks), but the warnings remained. Note that work on this MR led to "rediscovery" that we need to use TrackLinearExtrapolator when fitting Velo tracks, but this also does not help with the alignment warnings.

If interested, one can find the computation of smoother gain matrices in the new fitter here, the filtering step is done here.

Attached below you can find an example of stdout and log from 4th iteration of alignment sequences with both fitters ran with misaligned Velo alignment constants for input (3000 events, 4 iterations, based on Humboldt configuration of Velo alignment).

Templated AlignAlgorithm with TrackMasterFitter: alignlog_templated_tmf.txt, stdout_templated_tmf.txt

Templated AlignAlgorithm with PrKalmanFilter: alignlog_templated_prkf.txt, stdout_templated_prkf.txt

Alignment constants comparison from another run on misaligned Velo can be found in this pdf: alignment_constants.pdf For this reference nodes were turned off and parametrized scatter were used in TrackMasterFitter to make it more similar to PrKalmanFilter.

If interested, see option dump files from 4th iter here: opts_tmf.log opts_prkf.log

@wouter, @ausachov, @freiss, @sborghi