Unnecessary merge attempts

[kondziu]

Whenever multiple subsets of a dataset are used, a merge happens. The merged data is cached, so often the merge, finds nothing to do in the end, but it still takes time to get through the data. Find a way to skip merge attempts when we know a priori that nothing has changed. One idea is to store some sort of checksum-like metric that can be quickly calculated for the pre-merged subsets and checked before attempting to merge, on the theory, that if these are correct, the currently cached merged dataset can be used and the process of merging is skipped entirely.

Related to #6.