Connection.prefetch() could handle multi-databases more intelligently

[jamadden]

Right now, in a multi-database, if a user has an iterable of ghost objects, it's up to them to split it up by database so that it gets to the right connection to prefetch:

# sketch
iterable_of_ghosts = get_some_ghosts()
ghost_by_connection = defaultdict(list)
for ghost in iterable_of_ghosts:
    ghost_by_connection[ghost._p_jar].append(ghost)
for conn, ghosts in ghost_by_connection.items():
   conn.prefetch(ghosts)

If they don't do that, the results are anything between "only some data got prefetched" to "it was a bunch of pointless work, because this connection had none of them" to "this connection prefetched them, but they're not from this database, so it pushed a bunch of useful data out of the cache."

It seems like this could be better handled by the prefetch method itself.