Logging + Perf optimizations

[esemeniuc]

• Avoid relayer recomputing recipients on every packet batch (do every slot update instead
• Use Arcswap instead of rwlock for leader schedule (faster, no chance of starvation)
• Reduce datapoint rates, emit highest slot metric only when new slot
• Reduce channel forwarding for HealthManager
• Delete irrelevant metrics
• Less calls to fetch SystemTime
• Early return a bunch of places