Don't load all reference data values into memory

[LosD]

Currently. the reference data is always loaded into memory, needed or not. This could be an issue for some users with big/huge dictionaries.

DictionaryConnection is already returning an iterator, so it should be a simple matter of making a custom iterator implementation. However for SynonymCatalogConnection, we'll probably need to create a new method (and deprecate the old one), to do the same, then move all our own usages over.

[kaspersorensen]

A word of caution for this item: It was previously built so that they would not load into memory. And the consequence was that they where waaay too slow. So there is a definate trade-off here.

If anything, I'd suggest adding a configuration flag in the reference data's configuration. My impression is that huge reference data sets are rare and that small/medium size reference data (say, less than 10,000 entries) is much more common.