crc32_8 is a major performance bottleneck during bulk writes

The crc32_8 function is constraining performance of database writes, specifically, flushing the WAL. During a large fdb_commit that I've profiled, 32% of the time is spent in the crc32_8 function. (And the fdb_commit call was responsible for 70% of the time to write the database!)

I experimentally changed crc32_8 to call the adler32 function from libz, and saw a 20% improvement in commit time. (From the docs: "An Adler-32 checksum is almost as reliable as a CRC32 but can be computed much faster.")