feat: Allow model developers to specify MCP prompts for their models

[namabile]

Make prompts configurable at the model level using a json config file similar to what we're doing for publisher.json files.

Most MCP clients expose prompts as slash commands. Model developers could use this to expose common analyses that are specific to a model's use case in a standardized and repeatable way. For example, a sales model could expose a slash command like /sales:analyze_last_weeks_sales that provides a prompt to the client LLM telling it how to pull the data from the model using the MCP Server's tools, analytical steps the LLM should take based on the data, and how to format and expose the output to the client's user.