PR #1179 - Capability Reassessment Gets a Durable Cadence

The harness baseline is useful only while it is kept current. This change makes the existing synlynk capability sweep command part of a durable operational loop: state.db records the last successful sweep and the dispatch count at that point, while synlynk doctor and synlynk status warn after 25 jobs or 30 days.

The cadence check is deliberately observational. It does not run live probes or re-rate harnesses from CI. A maintainer explicitly runs the paid calibration sweep, and only a completed sweep advances the recorded timestamp.

Verification covers the never-run, under-threshold, and over-threshold cases with focused pytest runs. This closes the recurring reassessment gap tracked by #1179 while preserving the existing cost guardrail and independent verification flow of the sweep itself.