Determine if an issue with ngprecomputed on edge blocks still exists

[DocSavage]

    I was disappointed to see that the libdvid python bindings do not bind the `/specificblocks` endpoint, so I added support.  However, it's not returning what I expect.

For some reason the following call to DVID returns an empty file:

http://emdata6.int.janelia.org:9000/api/node/f3969/segmentation/specificblocks?blocks=493,426,770&compression=blocks&scale=0

And if I switch that to grayscale I get an error from dvid about the "minishard" from which the data is supposed to be fetched out of the neuroglancer precomputed volume.

http://emdata6.int.janelia.org:9000/api/node/f3969/grayscale/specificblocks?blocks=493,426,770

Error 400:

No minishard map found in shard "8x8x8/04ecb.shard" for minishard 3
 (/api/node/f3969/grayscale/specificblocks).

It smells like I'm doing something dumb, such as swapping X and Z. But as far as I can tell that's not the problem.

Originally posted by @stuarteberg in janelia-flyem/NeuTu#395 (comment)