Storage engine migration is recruiting two storage servers on same processes

[satherton]

The current migration logic intends to drop / add each storage engine one at a time without both old and new engines overlapping in lifetime, but testing shows this not to be the case.

This is particularly problematic since Redwood and SQLite do not share the same cache budget. While they could, in general it is hard to enforce that all storage engines share the same cache, for example RocksDB would also not share a cache budget with SQLite.