Currently, perfmon relies on reading log files created by ezmsg-sigproc.
Instead, we should look into whether we can use py-spy and parse its output to assign latency to individual graph components. Then if we dive into that component we can get a flame graph scoped to that component only.
