This change should reduce heap allocations (especially on the LOH for responses > 80KB) and the amount of bit blitting.

Note: unless we implement a custom memory pool, responses > 1MB will still be heap allocated. However, it should still reduce the number of heap allocations involved in this edge case, and a custom memory pool can be added in a future iteration if required.

Note: Follow-up with https://couchbasecloud.atlassian.net/browse/NCBC-1897#icft=NCBC-1897 is required to reach best efficiency.