[krbdev.mit.edu #3738] kdb db2 back end: lock per database, not global

[Ken Raeburn via RT]

Currently the kdb-db2 code has one big mutex for the whole plugin.

We should only do locking per Kerberos database.  If a KDC is using two
databases, they should be handled independently.  However, multiple
threads doing file locking on one database via different file
descriptors must coordinate.

This isn't important until the KDC and other KDC-side server programs
become multithreaded, and only if we find we encounter certain types of
performance issues.