Aggregate utilization expressed as fractional cores in use (0.0..n_cores).
Not clamped: kernel rounding can produce values very slightly above n_cores.
Per-process cumulative disk I/O bytes from /proc/pid/io.
Returns { pid -> (read_bytes, write_bytes) }.
PIDs whose /proc/pid/io is unreadable (e.g. different UID without ptrace)
are silently omitted โ the delta for those PIDs will be 0.
Returns a map of { pid to (utime, stime) } for every process in the tree
rooted at root_pid (root included). Processes that have already exited
are silently skipped: this is a TOCTOU race we accept.
Pure math: percentage of non-idle ticks between two snapshots (0.0โ100.0
before any clamping). Takes raw pre-computed totals/idles so it can be
unit-tested without constructing a CpuTime (which has private fields).