But Wait, There's More!

Minor bug fixes to prevent undefined behavior on some architectures.

Final implementation

BRC only uses 32MB of ram to encode the BWT in parallel or single threaded, this new version has better logic unrolling and decodes about 10% faster with about 5% faster encoding than the first version. Encoding and decoding can be done with any number of threads up to 16 (to prevent over pressuring any AVX registers).