tests demonstrating a failure case of CF digests

[rhuffy]

During the review of #639, some concerns were raised about the potential for digest collisions when added the PageToken to the column family digest.

We should expect CFs that are not equal to have different digests, except for the unlucky hash collision.
After close inspection, it appears that the existing code is actually vulnerable to collisions of crafted Cells or DeletionInfos.

A Cell digest has the following structure

[ name, value, 8-bytes-timestamp, 0]

A DeletionInfo digest has the following structure (assuming there is no row tombstone)

[ start, end, 8-bytes-timestamp]

name, value in the Cell and start, end in the DeletionInfo are arbitrary-length ByteBuffers, so the structure of these digests is almost identical, except for the suffixed 0 at the end of the Cell digest. With some simple bit shifting, we can easily craft a Cell and a DeletionInfo that have the same digest.

cell("A", "B", 1)
[ name, value, 8-bytes-timestamp, 0]
[ 65,   66,    0,0,0,0,0,0,0,1,   0]

tombstone("AB", "\0", 1 << 8)
[ start,     end, 8-bytes-timestamp]
[ 65, 66,    0,   0,0,0,0,0,0,1,0 ]