dsh-model-capabilities: Model Input and Inference Configuration

The lylarcher/dsh-model-capabilities plugin addresses the issue of some model configuration options being hidden in the default settings of DeepSeek Harness (DSH). It is available in DSH Web and Desktop and provides a settings page that allows users to directly enumerate all configurations, toggle input modalities (text/image), and configure Reasoning Efforts. Users must first configure the provider on the Models page; the plugin writes directly to the `llm-pi-ai` settings namespace and takes effect immediately. Although there are design limitations, the plugin provides a necessary tool for users who need fine-grained control over inference modes.