Change "bucket" algorithm

[jechols]

Changes the system to use MD5 to produce base-16 values, instead of relying on IDs directly. Using MD5 ensures very random-like buckets for better distribution of files across buckets, and in a perfect world, you won't see very many directories/files in any given "bucket" until you have a lot of files.

The existing system, since it uses just the id as-is, had a few concerns for us:

• It'll put most assets into the same few directories, especially with alphanumeric noids. Testing 1000 assets, we ended up with all of them showing up in 25/14/nX, where "X" is one of about ten letters. Further testing suggests it would take a long time before a new bucket at levels one or two would be created (the "25" or "14" in /25/14/nX).
• If the id is 8 digits or fewer, every asset will live in its own directory - probably not a real problem, but it seems like a high directory-to-file ratio
• If the id is extremely long (as an example, 14 digits), way too many assets will live in the same directory: 12345678901234 turns into /12/34/56/78 - and that directory will hold 12345678000000 through 12345678999999, or nearly a million assets.

I'm sure this approach is not the best way to do it, but at the least, a configurable "treeifier" seems like something worthwhile.

[coveralls]

Coverage increased (+0.2%) to 97.3% when pulling 389638a on jechols:feature/fix-buckets into 5eb5617 on projecthydra-labs:master.

[jechols]

Oops, my comment above is wrong - I used base-16 not 36.

[mjgiarlo]

Edited your comment to reflect that, @jechols

[mjgiarlo]

This is excellent. Thanks!