fix #1322
partially fix #1297

The post_exchangeSplit was originally conceived as a means of scaling duplicate suppression beyond what a single process could support. Duplicates were defined as files that had the same checksum, so splitting them among exchanges based on the checksum is what made sense initially. This method of load distribution will assign the same file path to different exchanges if they are not identical (don't have the same checksum.)

As work with mirroring has proceeded, it turns out that a file is really a stream of changes to a given path, and to make good decisions about it, you want a single process to receive all the updates for a given path. if a file is modified and then removed... the removal event would not necessarily have a related checksum to the modification. Similarly for a symbolic link.

Changing the hasing algorithm to be based on path name, rather than checksum:

• If files have the same path, and the same checksum... they are still duplicates. so that still works.
• If files have the same path, but different checksums... fine it should be processed same as today, just by the same participant in the exchangeSplit pool of flows.
• If files have the same path but different type (link, directory, file... remove event.) they should now show up in the same participant in the exchangeSplit pool of flows.
• it was a sorry that when hashing based on checksums, we hamper the effectiveness of the duplicate suppresion path by removing some hashes from it, resulting in an unbalanced tree. hashing on path is likely to yield better/faster associative arrays for duplicate suppression within each participant in the exchangeSplit pool of flows.