Investigate performance impact of using unicase sorting for sfld column

I suspect the impact might be quite heavy in fields with a lot of text - would be interesting to see before/after timings for a collection with sort fields of 20 bytes, 4k, and 1 megabyte apiece. If the performance impact is too large, we could potentially work around it by adding a new index, but since it would need to be added/removed when moving to schema 11, it's not something that is practical to do just yet.