storage: aggregate compaction_ratio metric across partitions

Although it is a bit more natural to leave this unaggregated to determine
the compaction ratio of a log/partition rather than a topic, the cardinality
of this metric is much too high, and its usefulness does not justify the cost.

Because this is aggregated, it is now a sum, meaning to recover the average
compaction ratio for a topic, we must divide by the number of partitions in
Grafana, e.g.

```
sum(compaction_ratio) by (topic) / count(compaction_ratio) by (topic)
```

Author: WillemKauf

src/v/storage/probe.cc

@@ -255,21 +255,17 @@ void probe::setup_metrics(const model::ntp& ntp) {
             "Number of segments that have been compacted away "
             "during adjacent merge compaction."),
           labels),
-      },
-      {},
-      {sm::shard_label, metrics::partition_label});
-
-    _metrics.add_group(
-      group_name,
-      {
-        // compaction_ratio cannot easily be aggregated since aggregation always
-        // sums values and sum is nonsensical for a compaction ratio
         sm::make_total_bytes(
-          "compaction_ratio",
+          "compaction_ratio_total",
           [this] { return _compaction_ratio; },
-          sm::description("Average segment compaction ratio"),
+          sm::description(
+            "Sum of segment compaction ratios across all partitions in a "
+            "topic. Divide by the number of partitions for a topic to get the "
+            "average compaction ratio per partition."),
           labels),
-      });
+      },
+      {},
+      {sm::shard_label, metrics::partition_label});
 }
 
 void probe::add_initial_segment(const segment& s) {