feat: implement MCP connection keepalive logic

[szmania]

Purpose

Implement a keepalive mechanism to maintain active MCP connections during periods of inactivity, preventing connection errors when tools are subsequently called.

Design Considerations

• Background Task: Implemented as an asyncio.Task within the HttpBasedMcpServer class.
• Heartbeat Mechanism: Uses lightweight OPTIONS requests to check server availability.
• State Machine: Manages connection health through CONNECTED, UNHEALTHY, and DISCONNECTED states.
• Resilience: Implements exponential backoff with jitter for automatic reconnection.
• Configurability: Supports a configurable keepalive_interval in the server configuration.
• Security: Includes jitter to prevent timing analysis and supports optional authentication headers.

Changes Made

• Added ConnectionState enum to cecli/mcp/server.py.
• Implemented start_keepalive(), _keepalive_loop(), and reconnect() in HttpBasedMcpServer.
• Updated connect() and disconnect() to manage the keepalive task lifecycle.
• Added validation for keepalive_interval configuration.

Testing Performed

• Unit Tests: Verified state transitions and reconnection logic.
• Integration Tests: Validated heartbeat frequency and connection maintenance with a mock server.
• Resilience Tests: Simulated network failures and verified automatic recovery.
• Concurrency Tests: Ensured the keepalive task does not interfere with other operations and cleans up correctly on disconnect.
• Configuration Tests: Verified validation of interval ranges and types.

Refer to .cecli.plans.md for full implementation and testing details.