[CBM] Backing up to cloud should reuse buffers to lower GC churn

What is the problem?
In a customer case we saw very high memory usage when backing up to S3 with lots of threads and lots of KV nodes. After some investigation we found that when there are a lot of vBuckets streamed at once to S3 the memory profile becomes dominated by riftAllocateObjstoreBuffer and that GC starts freeing a lot of garbage.

Further investigation was inconclusive but I have a feeling we are putting a lot of pressure on the GC with these buffers.

What is the solution?
We should see whether reusing the buffers improves the memory usage. This could be done using a Pool