Monitor and handle longer running requests

[johnml1135]

These can be found with the following Prometheus query:
histogram_quantile(0.95, sum(rate(http_server_request_duration_seconds_bucket{namespace="serval",http_route!~"(.*current-build|.*builds/{buildId})"}[2m])) by (le, http_route))

This should be able to be used to identify long running queries when they happen (especially if an alert is placed, such as expression > 4. The question is, what to do then?

Some of the issues raised first appear to be about word graphs - #489, but others are long running insert pretranslations (makes sense) or get corpus? Likely there is some other function that is running a long time that is holding up the database access. This can really only be effectively investigated after tracing (#138) is implemented.

[johnml1135]

After these things are addressed, we can add the prometheus alert to verify that the solutions are working and to proactively be alerted that there is a long running http request that we should investigate the trace of.