Describe the feature

We've discussed this a few times in slack but it seems worth having an issue for it and I didn't spot one, so here one is :).

As of right now we spawn pod scrapes in parallel for ~every revision with no real bound on the amount we'll attempt at once across revisions. This means we can potentially have an extremely large number of active outgoing scrapes, causing more contention than needed, exceeding keep-alive connection pool size and buffer pools etc. In our larger environments, we see large numbers of connections waiting on DNS and GC when revisions*pods gets large.

We should probably introduce some sort of work pool for the scrapers.