Add a route reconciliation step to the host_route_sync file that uses
the BGP state table as a system of record for the routes that should or
should not be synced to the system.

In theory, kube-router should be consistently informed of route changes
via BGP Path change events that are sent to us from GoBGP. However, in
practice it seems that some routes are getting stuck in our state table.

Host route syncing is an important part of kube-router that is run in
its own goroutine, so we should be checking it for health the way we do
the major controllers to make sure that it never stops functioning.

Additionally, as the health check controller has been continuously added
on to, we also make some modifications here to make it a bit more robust
and scalable.

In addition to added / removed routes, there is also the chance that the
gateway has changed and we need to track that as well by fixing the
cache state.

…tion

We don't want to add paths that are either not the Best path, which is
something that we get for free on the BGP Watch method that we have in
the NRC, or paths that are currently being withdrawn (even though this
should be an edge case).

In a List operation, we get our own routes that we've advertised. This
was a problem that we didn't have with BGP Watches. However, this now
affects us, so we need to remove IPs that represent ourselves, so that
we don't add needless routes.

Adding a boolean to the return values of InjectRoute() allows us to
track when a route is actually inserted vs. when it was called. This
helps us better represent HostRoutesStaleAddedCounter metric.

Without this, the administrator will see a bunch of routes seemingly
being added every time we call checkCacheAgainstBGP() when, in fact,
most of them are tossed out depending on the user's mesh subnet
settings.

There are a couple of items that have typically ended up in a no-op for
us when considering routes to inject. However, now that we have a route
map where we track route state, we need this not just to be no-ops, but
also update the route state cache as well to ensure that the route
doesn't get replaced in the future.

When we find tunnels to clean up, we need to not only remove the tunnel
and the route to that tunnel, but also remove the route from the state
map.

When we discover that no route needs to be added to the host because
it's not in the same subnet and we weren't supposed to create a tunnel,
then we also clean it up and ensure that it isn't in our state as well.