pstack 0.9.64 Adds Fresh Agents, Benchmark Checks

This is a disciplined reliability release: less flashy than a new model, but more consequential for teams asking coding agents to do repeatable engineering work.

• –Fresh agents with consolidated briefs reduce stale context and cross-task contamination.
• –Benchmark checks make performance claims an explicit verification step rather than an assumption.
• –Hourly audits reduce operational churn while requiring owners to leave pushed branches and decision trails.
• –The release grows pstack to 32 public skills and 24 principles across multiple agent harnesses.
• –Changes to PR tooling, autonomy defaults, and playbook sizing improve day-to-day operator clarity.