Enable bkw search for non-quadruplet pixel-seeded iterations, and update to latest hit selection windows

[mmasciov]

PR description:

As per title (in answer to BTV results showing issues with number of missing inner hits).

PR validation:

In 112X, with default 6 mkFit iterations:

• TTbar (PU=50): http://xrd-cache-1.t2.ucsd.edu/matevz/PKF/MTV_mkFit-6iter_ttbarpu50_112X/
• 10mu, 0.2<pT<1.0 GeV: http://xrd-cache-1.t2.ucsd.edu/matevz/PKF/MTV_mkFit-6iter_10muvlowpt_112X/
• 10mu, 0.2<pT<10 GeV: http://xrd-cache-1.t2.ucsd.edu/matevz/PKF/MTV_mkFit-6iter_10mulowpt_112X/
• 10mu, 0.2<pT<1000 GeV: http://xrd-cache-1.t2.ucsd.edu/matevz/PKF/MTV_mkFit-6iter_10muhighpt_112X/

Including tobTecStep (6+1 iterations):

• 10mu cube: http://xrd-cache-1.t2.ucsd.edu/matevz/PKF/MTV_mkFit-7iter_10mucube50_112X/

--> At built-track level, efficiency improves in all samples, especially at low pT. --> OOB performance: improvement in efficiency, with slight increase in fakes+duplicates. --> Number of pixel layers/hits increases in iterations where backward search is (now) activated. --> Slight reduction (< 0.5 hits) in number of strip layers/hits.