db: support arbitrary size write batches about pebble HOT 1 CLOSED

Another option: if a write batch is larger than the memtable size, we could index the write batch (if it isn't already) and add it as a memtable layer and flush it to disk immediately. This would avoid copying the memory, while not blocking the commit pipeline for anything other than the large write batch.

Rather than using the normal batch indexing mechanism, we could allocate an array of offsets and sort the array into key-order. This would use less memory than indexing the batch using a batchskl.

from pebble.