Reduce Prometheus Data Retention Duration

Nobody has claimed this yet.

Following a [recent slack report](https://linkerd.slack.com/archives/C89RTCWJF/p1580228200108300) on a Prometheus OOMKilled incident, we want to revisit the effect of reducing the `storage.tsdb.retention.time` duration from the current default of 6 hours to 1 hour. The previous investigation, per https://github.com/linkerd/linkerd2/issues/3258#issuecomment-526377964, showed that it did help to reduce the memory footprint. The obvious tradeoff is that there will be less metrics data available to the user.

It was also reported that reducing the `storage.tsdb.min-block-duration` to 30 mins and `storage.tsdb.max-block-duration` to 1 hour, from the default 2 hours, reduced Prometheus' memory usage by less than half. Note that these are [hidden flags](https://github.com/prometheus/prometheus/blob/9c67fce6e0123e3acc85c6ec4cc89e79cb2010d8/cmd/prometheus/main.go#L186-L188), marked for "testing only". We may or may not want to rely on them, but it's still worth investigating.

Open Questions:

* Shall we introduce make these settings configurable via CLI flags, config file (see https://github.com/linkerd/linkerd2/issues/3870), K8s configmap etc.?

Start by reviewing the Prometheus flags in cmd/prometheus/main.go linked by the issue, then read the related discussions in issues #3258 and #3870. Determine whether and where Linkerd should expose retention and block-duration settings, and define completion around a decided configuration path, validated defaults, and the documented metrics-retention tradeoff.

Written by the indexing model from the issue text.