Add 'kernel_restarts' prometheus metric

[yuvipanda]

"My kernel restarted, why?" is one of the most common support questions when supporting JupyterHubs. Being able to tell when a kernel was restarted - ideally being able to differentiate between user initiated restarts and nanny initiated restarts - would be extremely helpful to JupyterHub admins in providing this support, as well as detecting when memory limits and guarantees need to be changed (because too many users are running into memory limits)

[minrk]

Should the metric include the kernel ID? It might be useful if you want to identify a specific kernel that's restarting a lot (e.g. correlating it with logs), but I'm not sure that's usually useful.

[yuvipanda]

Let's just start without, and we can add them on if necessary? I can't think of a use for them right now though.

[minrk]

#1241 should do it. I'd love it if you have strong opinions on label names and enum values!