Fix logic for deleting Services from ProviderResources

[chauhanshubham]

Description
The current logic attempts to Delete Services from the ProviderResources map when no Routes are found on the cluster, for a given namespace.

gateway/internal/provider/kubernetes/httproute.go

Line 284 in 81eb817

r.resources.Services.Delete(request.NamespacedName)

Snippet

// Delete the Namespace and Service from the resource maps if no other
// routes exist in the namespace.
routeList = &gwapiv1b1.HTTPRouteList{}
if err := r.client.List(ctx, routeList, &client.ListOptions{Namespace: request.Namespace}); err != nil {
	return reconcile.Result{}, fmt.Errorf("error listing httproutes")
}
if len(routeList.Items) == 0 {
	r.resources.Namespaces.Delete(request.Namespace)
	log.Info("deleted namespace from resource map")
	r.resources.Services.Delete(request.NamespacedName)
	log.Info("deleted service from resource map")
}

This tries deleting a Service with namespace/name that matches the HTTPRoute which came in to be Reconciled. This will not be the case most of the times.
We must delete the Services which were referred to by the HTTPRoute that has been now deleted.

[danehans]

xref #539

[chauhanshubham]

Does having a separate cache for these mappings make sense? Something like

type ProviderCache struct {
   lock
   routeToServicesMappings map[string][]string
   serviceToRoutesMappings map[string][]string // *not having this still helps
}

Or we'll probably have to iterate through the ProviderResource cache multiple times
once to find out which services were referred by the deleted route
and second time to find whether the services are referred by other routes currently present.
^ both the steps must be performed for every kind of route in the ProviderResources