This makes it much easier to generate bindings for rhythmdb, and makes
creating custom entry types a bit more sane too.  Now python entry types
are created by subclassing rhythmdb.EntryType and overriding methods.
In C code, entry types can be created either using the old method of setting
function pointers or by subclassing.

RHYTHMDB_ENTRY_TYPE_INVALID no longer exists.  NULL indicates the absence of
an entry type.

The 'type' entry property is now a GObject rather than a pointer, so
rhythmdb_entry_get_pointer becomes rhythmdb_entry_get_object.

Entry types still cannot be unregistered, so reference counting of entry types
is not much of an issue.