This fixes an awesome bug where restoring from a pipe would not work,
because that's the case where you're going to see short Read()s.

The idea is to support a set of metadata checks of various
kinds, to select books in large databases for further processing.

The bleve API has changed slightly, too.

There aren't any external tools to backup a LevelDB database, so it's
nice to have a way to do so safely: the dump method is exposed to
localhost on the web interface.

Use a separate, simpler struct for indexing. Store unique IDs in the
index directly instead of parsing them back into a Metadata object.
Add more thorough tests.