When index processes are started, they can't tell if vbucket databases were recreated

Right now when an index process starts, it can't tell if the vbucket databases were recreated.
If any vbucket database was recreated, all the index entries that originated from that vbucket database have to be removed from the index and the index must index all changes (from sequence number 0) from the new database.

When index processes are alive, they detect database recreation, but when they aren't they can't know that - for example, the server was down and the database files were overwritten with backups, the index process was not yet started by ns_server and before it got started one or more vbucket databases were recreated, or, the index process crashed and before it got restarted some vbucket databases were recreated.

A possible solution here is to store a UUID/signature in database headers. Every time a database is created, it gets an UUID/signature added to its header. Index headers would store, besides the indexed seq number, the UUIDs/signatures of each vbucket database - when started, the index processes can compare the current database UUIDs/signatures with the ones stored in the index header to detect if the databases were recreated.

After talking to Damien, we don't know if the above solution is completely robust yet, but in principle there shouldn't be any use case that defeats it.
Needs some more thought however.