Some optimizations to the default StringConcatFactory might have made the code generation a bit more complex than necessary, and we can see that for more complex concat expressions we drop off in performance a bit too soon.

I've made a few consolidations and see improvements on multiple microbenchmarks with little to no change on bootstrapping overheads.

- Generate prependers that always append "" prefixes then remove the no-prefix prepend methods and simplify the code to do less branches and fewer operations on longs
- Same for the newArray combinator: always add a "" suffix
- Various other simplifications and optimizations