Various code simplifications, leading to speed improvements in HltControlFlowMgr

These changes are purely motivated by speed measurements and usage of flamegraphs. It looks like gcc has a hard time inlining/optimizing in the context of templated + recursive calls (here ControlFlowNode::reserved in particular).

It would be interesting to understand more deeply why these changes are helping.

As a rough number, we get a factor 2 speedup on TurboSprucing with this. And in theory, we could see a 5% gain on overall HLT2 (to be confirmed)