feat(benchmark): capture heap/allocs/cpu pprof profiles and gate Peak CPU on p95 (#934)

- Enable ENABLE_PROFILER on node-agent during benchmark runs and capture
  heap.pprof, allocs.pprof, and cpu.pprof per pod.
- Gate Peak CPU on 95th percentile (10% threshold) instead of strict max
  to eliminate false positives caused by sampling noise and single-sample jitter.
- Dynamically detect and wait for all operator-created node-agent daemonsets
  during install and image swap.
- Add documentation covering the gate methodology and pprof profile analysis.

Signed-off-by: Matthias Bertschy <matthias.bertschy@gmail.com>