GPU/TPC: Use qTot as marker for HIP clusters instead of qMax

[fweig]

No description provided.

[davidrohr]

I did 3 tests:

• Compared this version with hipTailFilter disabled to the O2 version before the hipTailFilter code in deterministic mode: 100% identical results
• Checked tracking efficiencies and ncl/track in this version with hipTailFilter enabled and disabled. Shows only marginal differences. And tiny differences are expected since the cluster will slightly differ.
• Run sync and async reco in the standalone benchmark and compared the saturated clusters, and they are identical. So they fully survive the track model compression / truncation.

What I did not do yet is to recompile O2 and create CTFs with it, but I am still traveling with only my laptop.

I also checked the dedx code, and it uses qtot directly, not the getter, but it also uses only 16bit int storage. So this needs to be rethought.

In any case, the cluster writing for saturated signals seems to be correct now, and the version is 100% backward-compatible.

[davidrohr]

errors unrelated, merging