Add MCP server support

This PR adds MCP server support to paperoni, enabling AI assistants and other MCP clients to interact with paperoni's paper search functionality.

Changes

• New MCP module (src/paperoni/mcp/):

  • client.py: HTTP client (PaperoniAPIClient) for interacting with paperoni's REST API
  • server.py: MCP server implementation using fastmcp that exposes paper search tools

• MCP tools exposed:

  • search_papers: Search papers with filters (title, author, venue, institution, dates, flags, semantic search)
  • count_papers: Count papers matching criteria without fetching results

• Configuration:

  • Added mcp section to config with API client settings
  • New MCP command in CLI to run the server (supports stdio and HTTP transports)

• Dependencies: Added fastmcp>=2.14.2

The MCP server can be run via paperoni mcp and supports both stdio (for MCP clients) and HTTP transports.