Improve NetworkTransport and add test coverage

[mattt]

In my experience, the existing NetworkTransport implementation hasn't been super reliable. In iMCP, users have observed random disconnections, connection hangs, and inconsistent behavior during network transitions.

This PR implements a robust connection management system to address these reliability issues:

• Heartbeat Mechanism ❤️: Regularly checks connection health to detect and recover from "zombie" connections ¹
• Automatic Reconnection: Implements smart reconnection with exponential backoff to handle temporary network issues
• Improved Error Handling: Better detection and recovery from various network failure scenarios
• Connection State Management: More reliable tracking of connection state and proper cleanup

This PR also adds test coverage to NetworkTransport, which is a big win. Getting that to work was a bit tedious, as it required defining an internal protocol to match the NWConnection (final class). But I suppose it wasn't any more onerous than the contortions we needed to mock the Foundation URL Loading System 🥲

Anyway, with these changes, local builds of iMCP are working much more consistently

Footnotes

1. I suppose we could use Ping for this, but it seemed easier to drop down from the protocol layer to the transport layer for liveness checks. ↩