Instrument graphql monitoring metrics

[focusaurus]

Look into how to expose operational metrics around the graphql service including:

• request rate overall
• request rate per query/mutation
• latency overall
• latency per query/mutation
• error rate overall
• error rate per query/mutation

[focusaurus]

I did a fairly brief/cursory npm search and I'd propose starting with epimetheus

[focusaurus]

Here's the prometheus metrics we get from this.

metrics.txt

[focusaurus]

So there's a technical gotcha with the express instrumentation in that we need the instrumentation middleware to be added to the express app as early as possible, and specifically before the apollo graphql middleware, in order for the instrumentation to properly track request start times and responses. At the moment, there's no way to do this. createApolloServer.js creates the express app and adds apollo middleware right away. We need to design a clean mechanism for this.

@aldeed suggested something along the lines of a new Reaction.registerPackage() option loadFirstExpressMiddleware: [] but we need to think through the right mechanics and names, etc.