Add /mcp restart command to respawn crashed MCP servers

Problem

When a stdio-based MCP server crashes or needs restarting during a Claude Code session, the session permanently loses access to that server's tools. There is no way to reconnect or respawn the MCP server without ending and restarting the entire Claude Code session.

For users running multiple concurrent sessions (e.g., supervisor/worker patterns with 4+ tabs), this means closing and reopening everything just to fix one broken MCP connection.

Proposed Solution

Add a /mcp restart command (or similar) that:

1. Kills the dead/hung MCP server child process
2. Respawns it using the same config from settings
3. Re-establishes the stdio pipe
4. Makes the server's tools available again

Optional: /mcp restart <server-name> to target a specific server, /mcp status to see which servers are healthy.

Current Workaround

Save all session IDs, close all tabs, reopen them. Works but loses in-flight context and is disruptive.

Environment

• Windows 11 (MINGW64)
• Claude Code CLI with multiple concurrent sessions
• Affected MCP server: jcodemunch-mcp (stdio transport)