dsh-outcome-loop: DeepSeek Harness Task Outcome Ledger and Acceptance Plugin

The `dsh-outcome-loop` plugin replaces subjective LLM judgments with mechanical evidence, addressing trust and audit issues in task orchestration for DeepSeek Harness (DSH). It provides a localized task outcome ledger with zero additional model cost, supporting acceptance criteria defined by mechanical standards such as tests, exit codes, and file states. It runs offline by default, ensuring users fully retain ownership and control over their data, and offers interactive features for task creation, adjudication, and export, making it suitable for scenarios requiring strict auditing and cost control.