Medium / long term ghost probability improvents

To collect the wanted / to be investigated improvements for ghost probability (in principle in HLT2)

• Going to chi2/dof instead of chi2s for VELO for more inclusive approach (might lose some discrimination power)
• UT-segment chi2s
• UT geometric acceptance (not penalize tracks out of acceptance for having no UT hits, ...)
• Similar considerations for FT? Or VP?

fyi: @tskwarni @sblusk @decianm