performance improvements and benchmark for this library

[cdepillabout]

Performance has been almost completely ignored in the Data.Char.Unicode.Internal module. There are TODOs at the bottom of the file with particularly bad examples of this. It would be nice to fix all these TODOs.

It would also be nice to have proper benchmarks for this library.

[cdepillabout]

I believe the worst of the performance problems have been solved in #14.

It would still be nice to have a benchmark.