Add Jupyter Events and emit events from the Kernel Manager

[Zsailer]

This requires a release of Jupyter Events... done 👍

• Adds an event schema for the KernelManager
• Adds events to various methods in the default KernelManager implementation.
• Adds tests to the kernelmanager's lifecycle unit test to ensure events are emmitted.

[ccordoba12]

This looks like a major change, so hopefully for Jupyter-client 8, right?

[Zsailer]

@ccordoba12, yeah, I think that's fair. This adds an additional dependency, jupyter_events, which I think warrants a major release.

[blink1073]

Good call @ccordoba12, I've opened #833

[kevin-bates]

This looks good @Zsailer - just have one question.

Is this in anticipation of supporting other actions, because all of the currently defined actions fire with a non-None kernel_id? If we retained that approach, we could then apply a format or pattern, allowing only UUID formats (and remove the oneOf and null). This might allow event consumers to have an easier experience as well.

[Zsailer]

because all of the currently defined actions fire with a non-None kernel_id?

I don't think start_kernel or pre_start_kernel methods strictly require a kernel_id to start, right? This logic seems to suggest that if None is provided, it creates a uuid on the fly.

[kevin-bates]

That is correct (on both counts; non-required kernel_id and auto-create logic). The provisioners should have access to the kernel-id from the get-go so they can do indexing, etc. as necessary.

[blink1073]

LGTM, thanks!