Small issues in block_radix_rank.h

[danpovey]

https://github.com/NVlabs/cub/blob/c3cceac115c072fb63df1836ff46d8c60d9eb304/cub/block/block_radix_rank.cuh#L353

This documentation is quite unclear: ///< [out] For each key, the local rank within the tile

E.g. does it mean the number of previous keys with the same value or does it mean the number of distinct keys with values less than this key?

Also here https://github.com/NVlabs/cub/blob/c3cceac115c072fb63df1836ff46d8c60d9eb304/cub/block/block_radix_rank.cuh#L385

            thread_prefixes[ITEM] = *digit_counters[ITEM];

I have a hard time seeing why this value loaded would be nonzero, since the digit counters had just been zeroed. Sorry if I'm totally wrong about this.