If the btree uses prefix compression, calling next_raw_key_value_batch should allocate a buffer owned the cursor, where it reconstructs all keys that were encoded with prefix compression. The WT_ITEMs in the output key array should point to the buffer and/or the page itself as appropriate.
The function may reuse the buffer that the cursor already uses for returning the next prefix-compressed key via the standard next/get_key interface.
We should also include the corresponding functionality tests.