Speed up Moore initialization (XML reading)

Avoid using DOMNodeList::getLength() for a speedup of about x2 (or 30 s) in the XML reading for loading geometry/conditions. This dominates the initialization, which is sped up correspondingly.

More (sligthly less trivial) optimizations are possible.

This is very old code that is going to be made obsolete with Detector/DD4hep. However, initialization takes a considerable time when testing both centrally and locally (we have >100 tests in Moore), so it is worth improving that now.

🎉 Comparing the reference and the test build shows a nice little improvement from 2h30 down to 1h56 for the Moore tests job (for the v3 platform)