Snappy should have been explicitly included in the runtime, for I
erroneously thought that LevelDB bundled this into its runtime as-is.
It turns out that this assumption is wrong, and I thought we had
Snappy compression support all-along.
Presently our use of LevelDB and its operating modes are hardcoded
into the storage stack. This pull request decouples this and
re-exposes this through flags. We can now perform benchmarking
and remedial tuning.