DeepSeek Harness Yogācāra Self-Model Plugin

`dsh-yogacara` is an agent self-model plugin designed for DeepSeek Harness (DSH). Based on the Yogācara framework, it maps agent behavior to the eight consciousnesses and fifty-one mental factors, perceiving the agent's internal state from data such as tool results and code state. The plugin injects computed mental factors (e.g., conceit and doubt) and memory seeds into the system prompt to facilitate the agent's self-regulation and transformation. Through fine-grained state management, it provides developers with an observable and debuggable view of the agent's internal state.