DeepSeek Harness Plugin: dsh-adaptive-effort

The `dsh-adaptive-effort` plugin in DeepSeek Harness (DSH) uses a MiniMax scoring mechanism to dynamically adjust the reasoning level (`reasoning_effort`) based on user message complexity, balancing cost and depth of thought. It integrates token accounting, report statistics, reply metadata tagging, and compatibility handling for GLM-5.3. It supports heuristic fallback and heavy-workload tooling upgrades, and provides CLI reports to help users verify consumption across effort tiers, preserving deep reasoning for complex problems while reducing costs on trivial tasks by lowering the effort level.