v0.18.0

Changed

• Performance pass across every crate, measured with new ignored timing tests
  (2000 synthetic processes):
  • Policies: one tick of rules, limits, kill watch, insights and the other
    policies costs about 2 ms instead of 55–70 ms. Wildcard matching no longer
    allocates, and per-rule state is kept only for processes a rule matches.
  • Window: the process table is filtered, grouped and sorted only when the
    data, search, filters or sort change, not on every frame; a frame of the
    Processes tab takes about 1 ms instead of 2–3.5 ms.
  • Service: snapshot, status and recent-event replies are encoded straight
    from shared state instead of copied first; history samples and queries
    allocate far less (History tab query 3–4 times faster); the start watcher
    runs only when a rule needs it.
  • System queries: GPU and disk counters, process snapshots, connection
    tables and the disk and network traces reuse their buffers and make fewer
    calls; per-event maps use a faster hasher.
  • Space analysis: workers share folders at every depth, so one huge folder
    no longer runs on a single thread (about twice as fast there).
  • Process tree lookups and sorting by name are 6–10 times faster.