fix: propagate server errors instead of leaking continuation

The withCheckedContinuation that never resumed caused two problems:
1. SWIFT TASK CONTINUATION MISUSE runtime warnings
2. If Hummingbird crashed in the background Task, the error was silently
   swallowed — the process stayed alive but stopped accepting connections,
   causing silent disconnects from MCP Router with no auto-recovery.

Now main.swift awaits transport.waitForTermination() which propagates
errors from the server Task. If the server dies, the process exits and
launchd restarts it, allowing MCP Router to reconnect.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: cyberpapiii

swift/Sources/iMessageMax/Server/HTTPTransport.swift

@@ -128,6 +128,16 @@ public actor HTTPTransport: Transport {
         }
     }
 
+    /// Blocks until the server terminates, propagating any errors.
+    ///
+    /// Call this from main.swift after `connect()`. If the Hummingbird server
+    /// crashes or encounters an error, it propagates here — causing the process
+    /// to exit so launchd can restart it.
+    public func waitForTermination() async throws {
+        guard let task = serverTask else { return }
+        try await task.value
+    }
+
     /// Handles POST requests with JSON-RPC messages
     private func handlePost(
         request: Request,

swift/Sources/iMessageMax/main.swift

@@ -56,9 +56,7 @@ struct iMessageMax: AsyncParsableCommand {
             )
 
             try await transport.connect()
-            await withCheckedContinuation { (_: CheckedContinuation<Void, Never>) in
-                // Server runs indefinitely until process is terminated
-            }
+            try await transport.waitForTermination()
         } else {
             // Stdio mode: Single Server instance managed by MCPServerWrapper
             let server = MCPServerWrapper()