DynamoDB "metrics" autoscaler enhancements

[bboreham]

Things left over from #841:

• protect against scaling-down when Prometheus restarts and has no data (the rate() divides by the whole time period so looks lower right after startup)
• expose some metrics - e.g. requested throughput
• more aggressive scale-up when the queue is really big
• automatically increase capacity when an ingester rollout is under way (or fix Mysterious flush of underutilised chunks 1hr after ingester rollout #467)
• support use of Cortex to supply the metrics

[bboreham]

I think the first item should be fixed by fetching a vector of actual points rather than a matrix of averaged points.

[stale]

This issue has been automatically marked as stale because it has not had any activity in the past 30 days. It will be closed in 7 days if no further activity occurs. Thank you for your contributions.