Clarify which of the libraries uses RAM or disk and can persist an index

[pombredanne]

There is one part that makes a huge difference in practice is if a library works only in RAM or can create an index ahead of time and run from a disk/db/mmap file.
If a library cannot do this, it may be great in theory and quite unusable in practice making the benchmark rather biased towards better/faster algos that may not have a practical application at scale.