gracefully handle ratelimits

[calvinfo]

The ECS API throttles pretty aggressively. Right now we just throw when we try and hit it with a ton of concurrency, but ideally we would limit the concurrency and backoff in the face of errors.

[stephenmathieson]

this is something i was worried about with the polling. since we can't really handle the errors, we'll just keep serving old (potentially invalid) data :/

[larseen]

One way of dealing with ratelimits could be to rewrite the server logic to use the ECS Event Stream instead of polling for new data, thereby eliminating the problem entirely.
This would also improve usability since the data is always updated, with the newest data from the event stream.