Workaround issue with calculating length of computed resources

We have to pass the number of subnets around (rather than dynamically
counting them) because of an issue in Terraform:

hashicorp/terraform#12570

Without this we see the following error message when `terraform
apply`ing from scratch:

```
Error: Error refreshing state: 1 error(s) occurred:

* module.hsm.data.aws_subnet.vpc: data.aws_subnet.vpc: value of 'count'
cannot be computed
```

If the cluster has already been spun up and the HSM module included at a
later date then the Terraform would apply cleanly in its current state.

This may be end up being resolved in Terraform 0.12 when it is released.

modules/gsp-cluster/outputs.tf

@@ -22,6 +22,11 @@ output "private-subnet-ids" {
   value = ["${aws_subnet.cluster-private.*.id}"]
 }
 
+// Workaround for https://github.com/hashicorp/terraform/issues/12570
+output "private-subnet-count" {
+  value = "${length(data.aws_availability_zones.all.names)}"
+}
+
 output "controller-instance-profile-name" {
   value = "${module.k8s-cluster.controller-instance-profile-name}"
 }