Character classes for Unicode-aware lexers and regex engines

A faster, smaller Set.Make(Uchar). Cost follows the run count, not the cardinal: the Alphabetic property is 147,421 codepoints in 761 runs, held in 12KB against 5.6MB, and the whole codespace is 40 bytes against 42.4MB. Alongside the usual set algebra there is a builder for accumulating a class from many fragments, a compiled two-level bitmap trie for membership in an inner loop, partition refinement for derivative classes during DFA construction, and a stable packed encoding for embedding generated tables in source. Surrogates are excluded by construction.