and swapmod, where we use -X (syncVlansFromTables) to update existing
vlans:

The basic problem to solve is leaving vlans (and their lanid) behind on
switches that later interfere with a new lanid that uses the same tag.

Consider: when we create a new vlan, we operate on only the devices
needed to create that vlan (the stack object is created with those
devices). Then we reserve a vlan tag, asking the stack for a number that
does not exist on any devices in the stack. But it won't know about a
stale vlan with that number on some other device. Later we add more
ports to the shared vlan, on different devices (which includes the
interswitch trunk links).  We do not check to see if that vlan is
already on some of those devices with a different or no name, We just
plow ahead. What happens is somewhat switch dependent, but usually
something happens without an error.

And the main reason for leaving these behind, is that while we remove
the vlans from ports on the interswitch links, we do not remove the vlan
from the switches. Next time we come along and modify or delete the
vlan, the stack might not include those switches.

So ... the main effect of this commit is to try much harder in
setVlanOnSwitchTrunks(), to remove vlans from switches that no longer
have any ports in that vlan. We shall see how well this works ...

Note that shared vlans do not store the current switch path in the DB,
which is a handy debugging item. This commit changes that; we know store
the switch path in the DB, although in order to do that, we need to
serialize snmpit on shared vlans. This is seems fine, we do not do that
many shared vlans.