Define bounded CPU performance evidence

[itdevwu]

Goal

Define bounded sampled CPU evidence as a contract distinct from endpoint latency and exact aggregate duration, then prove it with a minimal PID-scoped collector.

Acceptance criteria

• Public schemas distinguish event-pair latency, exact aggregate duration, interval counters, and statistical samples.
• Sampling frequency or period, duration, stack depth, sample capacity, group capacity, multiplexing, and loss are explicit.
• Validation reports perf permissions and target identity without attaching.
• Sampled activity is never described as exact causality and cannot silently enter the existing correlator.
• One minimal collector and deterministic fixture prove capacity, cancellation, target reuse, malformed input, and cleanup semantics.
• The public CLI remains doctor, discover, validate, and measure; no daemon or persistent session is introduced.