Address more review comments to the vertex finding code

[fwyzard]

Use std::clamp(...) in device code now that CUDA supports c++17. Name reused constants in the vertex fitting and splitting.

[fwyzard]

No impact on the computing performance running the pixel-only profiling workflow.

Before:

Running 3 times over 10100 events with 2 jobs, each with 10 threads, 10 streams and 1 GPUs
  1339.5 ±   0.3 ev/s (10000 events, 99.4% overlap)
  1334.1 ±   0.3 ev/s (10000 events, 99.8% overlap)
  1328.4 ±   0.3 ev/s (10000 events, 99.1% overlap)
 --------------------
  1334.0 ±   5.6 ev/s

After:

Running 3 times over 10100 events with 2 jobs, each with 10 threads, 10 streams and 1 GPUs
  1342.0 ±   0.3 ev/s (10000 events, 99.6% overlap)
  1338.0 ±   0.3 ev/s (10000 events, 99.5% overlap)
  1332.8 ±   0.3 ev/s (10000 events, 99.6% overlap)
 --------------------
  1337.6 ±   4.6 ev/s