Add support for pending kernels

[blink1073]

Add MultiKernelManager.use_pending_kernels to use a created kernel before the process has started, based on original discussion in jupyter_server.

Add a KernelManager.ready property to that resolves when the process has started for the first time.

Refuse to restart a kernel that has not been fully started - we could also decide to wait if it has been launched but not fully started.

Consumers using use_pending_kernels should wait for the ready future before attempting to communicate with the kernel.

This change will correspond to a change in jupyter_server, where the server will use the ready future before connecting the websocket.

Web clients will have to be updated to handle errors on WebSocket startup and surface them to the user.

As a result, the use of use_pending_kernels will be opt-in for now.

• Add MultiKernelManager.use_pending_kernels
• Add KernelManager.ready
• Make sure a kernel that died at startup can be handled in MultiKernelManager.shutdown_all
• Add test for bad kernelspec causing startup errors
• Make sure existing tests pass
• Verify functionality in jupyter_server - Use pending kernels jupyter-server/jupyter_server#593
• Verify functionality in JupyterLab - Fix Handling of WebSocket Startup Errors jupyterlab/jupyterlab#11358

[blink1073]

Thanks @davidbrochart. I marked it as draft for discussion and because I still need to add tests for the new feature.

[blink1073]

TODO:

• Add a test where there is a bad kernelspec.

[kevin-bates]

LGTM - thanks Steve!

[blink1073]

Thanks for the review! I'll make a release with this on Monday