Collect bounded PID-scoped CPU stack inventory

[itdevwu]

Goal

Give an agent a broad, bounded on-CPU view of an unknown host workload before it chooses exact event selectors.

Acceptance criteria

• measure exposes sampled CPU inventory without adding a public command or routing samples through endpoint correlation.
• Collection is scoped to one PID/start-time identity and bounds frequency or period, duration, stack depth, retained samples, and groups.
• User/native stacks are symbolized from target mappings using build identity plus file-relative addresses; unresolved frames remain explicit addresses.
• Output reports inclusive/exclusive counts, sample coverage, lost samples, stack truncation, capacity utilization, and deterministic ordering.
• Inventory emits evidence-based native selector hints only when an attachable file-backed symbol is resolved.
• Live tests and benchmarks cover CPU-bound native/Python work, permissions, target exit/reuse, cancellation, capacity, and cleanup.