Make WEBrick test teardown block until the server thread exits

[Copilot]

The flaky HTTP output specs were starting WEBrick on a background thread and tearing it down asynchronously, which could leave the listener alive between examples and hang retries against a server that had not fully stopped.

• Return the WEBrick thread from the shared test helper

  • start_app_and_wait now returns both the server instance and the thread running app.start!
  • this gives teardown a reliable handle for waiting until WEBrick has actually exited

• Add a blocking teardown helper

  • introduced stop_app_and_wait(app, server, server_thread)
  • shutdown order is now:
    1. app.stop!
    2. server.shutdown
    3. server_thread.join(30)
  • if shutdown does not complete within the bound, the helper emits a clear warning and kills the thread instead of silently hanging

• Update the existing spec call sites

  • both the main HTTP spec setup and the TLS-specific setup now capture @server_thread
  • both teardown paths now use the shared blocking helper instead of manually shutting down WEBrick

@server, @server_thread = start_app_and_wait(TestApp)

stop_app_and_wait(TestApp, @server, @server_thread)

• Fixes Flakyness in tests due to WEBrick shutdown sequence #164