Cache Viewer Permission on Server Startup

As part of the epic to automate MCP server startup and reduce agent protocol (#7604), we should proactively fetch and cache the user's permission level for the repository when the github-workflow server starts. Currently, an agent would need to explicitly call the get_viewer_permission tool, which is an unnecessary extra step.

By fetching this information on startup, we simplify the agent's workflow and make the permission data readily available to all internal services.

Proposed Solution

1. Add a Cache Field: In ai/mcp/server/github-workflow/services/RepositoryService.mjs, add a new class field to the singleton, such as viewerPermission = null;, to hold the permission string.

2. Fetch on Startup: In ai/mcp/server/github-workflow/mcp-stdio.mjs, modify the main() startup function. After the initial health check, call the RepositoryService.getViewerPermission() method once and store its result in the new viewerPermission field on the RepositoryService singleton.

3. Refactor getViewerPermission Tool: Modify the getViewerPermission method in RepositoryService to be a simple, synchronous getter. Instead of making a GraphQL call, it should just return the value cached in the viewerPermission field. The actual GraphQL call will be moved to a new private method (e.g., #fetchViewerPermission) used only during startup.

Acceptance Criteria

• When the github-workflow server starts, it makes one call to determine the viewer's permission level.
• The permission level is stored on the RepositoryService singleton.
• The get_viewer_permission tool reads from the cached value and does not make a new API call.
• Other services can import RepositoryService and access the permission level directly (e.g., RepositoryService.viewerPermission).