Incoming reports clustering by similarity

[ksy36]

Changes in this PR:
import_reports_from_bigquery is only saving reports to db without cluster_id or bucket_id
triage_new_reports command gets reports that don't have bucket_id and attempts to cluster and bucket them (runs every hour at the moment)

I think once we import live reports the frequency of triaging can be increased.

[ksy36]

I need to find a way to run cluster_reports command once to cluster existing reports on production before these changes are deployed :)

[ksy36]

Ok I've added a page at /reportmanager/clustering/ where we can trigger full clustering and it also displays all runs, full and incremental ones (triaging new reports). We can probably display more info in the table later, if needed.

[jgraham]

I mainly reviewed the backend changes so far; for the frontend I wonder if we can get away without having this feature in the first instance.

[jgraham]

How important is this lock if we can't run stuff from the UI? I guess it still makes sense, but it also makes sense that we can't run multiple import jobs in parallel for example, in which case maybe we should have a more generic locking mechanism rather than one specific to this job type? I guess we're also using it to track progress, but again it seems like there's a lot in common with what we'd want to track progress of other jobs e.g. import. Not a change for this PR.

[ksy36]

Yeah, triage_new_reports runs every hour, so it's unlikely (but possible) that it will run at the same time as full reclustering (which I'm thinking to run with a command directly). Also on the first run triage_new_reports checks whether there are any successful runs, i.e. initial clustering exists and returns early if not. I guess it could also trigger initial clustering once (if it doesn't exist) without me manually running the command.
But yes I agree about generic locking mechanism.

[jgraham]

I think we should land this (once the lint is fixed!) so we can start to see the results and experiment with it.