Add metrics aggregation capabilities

[VladLazar]

This PR contains a set of patches from the seastar mailing list (patch set and repo). They add aggregations
support to the seastar metrics subsystem. This PR used to be based on an older version of the upstream patch set,
but Ben fed back our findings and the new version matches our needs as is.

Mailing List Cover Letter

Motivation:
Histograms are the Prometheus killer. They are big to send over the
network and are big to store in Prometheus.
Histograms' size is always an issue, but with Seastar, it becomes even trickier.
Typically, each shard would collect its histograms, so the overall data
is multiplied by the number of shards.

This series addresses the need to report quantile information like latency
without generating massive metrics reports.

A summary is a Prometheus metric type that holds a quantile summary (i.e.
p95, p99).

The downside of summaries is that they cannot be aggregated, which is
needed for a distributed system (i.e., calculate the p99 latency of a cluster).

The series adds four tools for Prometheus performance:

1. Add summaries.
2. Optionally, remove empty metrics. It's common to register metrics for
   optional services. It is now possible to mark those metrics as
   skip_when_empty and they will not be reported if they were never used.
3. Allow aggregating metrics. The most common case is reporting per-node
   metrics instead of per shard. For example, for multi-nodes quantile calculation,
   we need a per-node histogram. It is now possible to mark a registered
   metric for aggregation. The metrics layer will aggregate this
   metric based on a list of labels. (Typically, this will be by shard,
   but it could be any other combination of labels).
4. Reuse the stringstream instead of recreating an object on each
   iteration.

force push:

• split into smaller commits
• use sm::label strong type instead of std::string for specifying aggregation labels
• reuse string stream when converting aggregated metrics to text

force push:

• Remove aggregation_labels argument from metric creation functions (make_<metric_type>).

force push:

• Use the write_counter function in the commit it's introduced

force push:

• Use strong label type sm::label in setter of aggregation labels

force push:

• Amnon's latest version of the patch set includes everything we need and is functionally
  equivalent to what we had on this branch before (it also allows for skipping reporting of
  empty metrics configurably).

[BenPope]

This would be easier to review if the last commit was split up a bit and the comments improved.

[BenPope]

The patchset brought in optimises histograms by not outputting them if they aren't used. I wonder if we always want that. Maybe it should be configurable.

[VladLazar]

The patchset brought in optimises histograms by not outputting them if they aren't used. I wonder if we always want that. Maybe it should be configurable.

Didn't you remove that in this commit (the value.is_empty() check)?

[BenPope]

The patchset brought in optimises histograms by not outputting them if they aren't used. I wonder if we always want that. Maybe it should be configurable.

Didn't you remove that in this commit (the value.is_empty() check)?

Looks like it. But it was also printing the headers twice. Lets consider this later.

[VladLazar]

This would be easier to review if the last commit was split up a bit and the comments improved.

I split it up and expanded on the commit messages in this force push.

[BenPope]

I'm happy with this. Sent some minor nits to the mailing list.