Enhance MCP prompt argument handling with multi-word support and interactive discovery

Current Behavior

MCP server prompts in Claude Code have limited argument handling capabilities:

• Arguments must be single continuous strings (e.g., /create_task add_colors_to_palette)
• Multi-word arguments with spaces are not supported
• No argument discovery in the UI - autocomplete only shows command name and description
• Users cannot see what arguments a command accepts (arguments are never shown to users)

Problem

This limitation makes it difficult to:

1. Pass natural language arguments (e.g., task descriptions, commit messages)
2. Understand what arguments a command expects
3. Know if arguments are required or optional
4. Format arguments correctly on first attempt

Proposed Solution

1. Multi-word argument support: Allow quoted strings or other delimiters for multi-word arguments

• Example: /create_task "add more colors to palette"
• Example: /commit "feat: add user authentication"

1. Interactive argument discovery in the UI:

• Show argument list when a command is selected
• Display argument names, types, and descriptions
• Indicate required vs optional arguments
• Show example usage

1. Enhanced autocomplete that could show:

``
/create_task - Create a new task
Arguments:
title (required): Task title
priority (optional): low|medium|high
Example: /create_task "implement search feature" high
``

Use Case

Many MCP servers need to accept natural language input (task descriptions, search queries, commit messages) which currently require awkward underscore-based workarounds.

Additional Context

The MCP protocol already supports rich prompt definitions with argument schemas (MCP Prompts Spec). This request is about surfacing that capability in the Claude Code UI.