c/p_balancer: fixed possible overflow of final used ration

Replicas eviction policy is asynchronously applied to each partition it
may happen that the size of individual partitions is not exactly equal
to the total usage size reported. In this case the individual partition
sizes that are moved out of the node may sum up to a value larger than
the total used bytes. Added a code preventing overflowing the final
ratio in this case

Signed-off-by: Michał Maślanka <michal@redpanda.com>
(cherry picked from commit ba31bfb)

src/v/cluster/partition_balancer_types.h

@@ -41,6 +41,13 @@ struct node_disk_space {
     double peak_used_ratio() const { return double(used + assigned) / total; }
 
     double final_used_ratio() const {
+        // it sometimes may happen  that the partition replica size on one node
+        // is out of date with the total used size reported by a node space
+        // manager. This may lead to an overflow of final used ratio.
+        if (released >= used + assigned) {
+            return 0.0;
+        }
+
         return double(used + assigned - released) / total;
     }