Summary: Current folly version uses a single crc32c.  crc32c has a latency of 3, so pipelining three in a row makes it nearly 3x faster (for data all in cache).

Reviewed By: yfeldblum

Differential Revision: D5418228

fbshipit-source-id: d3a250e1b4fe1f0bc99b44c660df94cf233aebd6