TECH: StringDeduplicator: deduplicate only emptystring by default

[c247t]

Everything else might be too small to outweigh the overhead, or not even that redundant to begin with. We should provide some other means to decide on a field to field basis when to deduplicate inside de-externalization.

[kelunik]

Is this an assumption or baked by data?

[c247t]

It is backed by the fact that loading WatchDAO from dumps yields a DAO that looks smaller, but a StringDeduplicator that is much fatter, so totals are far worse. This is at least part of the reason why heap consumption keeps slowly creeping up over time.

[c247t]

The optimum solution would be to define the behavior through an annotation property, so that fields which profit greatly from content deduplication can make use of it again. For the general case though, we should enforce only scenarios where the win is guaranteed, as with redundant emptystring instances. Plus, for any temporary entities not kept in memory, deduplication means just a waste of additional allocation and gc.