When 'truncate' is true for an Indexer's constructor, by a
quirk of fate, it will always write a schema file named
"schema_1.json". If the existing schema file is also named that
(as it would be on successive truncating runs), the files will
collide, resulting in either a clobber a crash depending on OS.
The solution is to derive the name of the schema file from the
name of the snapshot file.
Original user list trouble report: http://s.apache.org/90L