Minimize number of random record file reads during Curly compaction

[thomas-kielbus]

While we perform the key file merge sort, we do random reads of records to be written from the record files being merged. It seems like there is a lot of potential to optimize this process. We could merge sort a fixed number of keys (e.g. 1024) and then only read once from each record file and get all the necessary records in memory before writing them to the compacted partition.

[thomas-kielbus]

It is important to note that there is no guarantee that the record file will be sorted. However, it will usually be the case and we should take advantage of that.