add compatibility requirement for propagations to strip hits

[slava77]

The following condition is applied on hit-prop residuals to make sure that the propagation points consistently within the length bounds of the strip

• strip hit half-length + 3*sigma_propagation > residual_along_strip

More detail:

• the strip half-length is related to the uncertainty of its position, in the simplest case of a hit aligned along z, the hit uncertainty along z is strip half-length times sqrt(3)
• in the barrel the test is simply done in z direction
• in the endcaps the test is done by projecting on the longest component of the hit correlation matrix (taking an approximation that in x-y the uncertainty across the strip (around 100 µm) is negligible.

This is a bit of a kludge until we get the proper geometry details. The CKF tracking has a similar requirement.

Validation using mkfit=all and compiled with AVX2:=1 USE_INTRINSICS:=-DMPT_SIZE=1

• ttbar http://uaf-10.t2.ucsd.edu/~slava77/figs/mic/mtv/ttbar50_mkFit-pr344-pr345-qCompat_m6787d91_c5e29902
• 10mu http://uaf-10.t2.ucsd.edu/~slava77/figs/mic/mtv/10mu_mkFit-pr344-pr345-qCompat_m6787d91_c5e29902/plots_initialStep.html
  • here, for 10mu, the plots show that there is no efficiency or hit loss

In the following, screenshots are for built tracks from the initialStep in the ttbar PU50 sample
orange is the same as the latest physics performance in this PR (6787d91) and has a small improvement compared to black, which was shown during the Aug 27 meeting

There is a small reduction in the number of hits due to rejection of fake hits

The number of missing outer is a bit better

a comparison with CKF is available in http://uaf-10.t2.ucsd.edu/~slava77/figs/mic/mtv/ttbar50_CKF_mkFit-pr344-pr345-qCompat_md94a476_c5e29902 (here qCompat is as of the earlier version shown in black in the mkfit=all comparisons above).

[slava77]

some time ago we talked about improving SelectHitIndices by making corrections to approximately estimate the propagated position with respect to the hit (φ, q). This PR is a post-propagation check of a similar kind in q.
Hopefully this can be eventually moved/reimplemented in the earlier stage in SelectHitIndices and save a bit in the propagation cost.