add ci-pr deployment

[cirocosta]

Hey,

This PR is intended to allow the prs pipeline to co-exist with
nci in the hush-house gke cluster.

it does so by:

• adding a node pool to the cluster (ci-workers-pr) whose size is the
  same as we had previously in our BOSH deployment
• adding a deployment (ci-pr) that puts workers in such node pool.

As such deployment is supposed to run untrusted workloads, we needed a
way of restricting the network access that this deployment could have in
order to avoid lateral movements in the internal net (in the case of our
BOSH environment, we had a totally different network - in k8s, we'd need
to be in a separate cluster if we wanted to go w/ the same approach of
different nets).

This led us to enabling the enforcement of network policies in the
cluster, and creating a policy for ci-pr that would target the pods
delpoyed by it, effectively blocking internal connectivity to anything
we didn't want (i.e., anything that's not ci's TSA).

ps.: these changes have already been applied.

related: concourse/prod#36

[cirocosta]

merging it directly to reflect the current state

[cirocosta]

w/ regards to "why to go w/ internal connection for connecting to tsa: it turns out that calico (I guess?) puts an external constraint on communications to even external services:

# matches tcp conns to our external load-balancer - once matched, gets us to the
# `KUBE-FW-blabla` rule.
#
-A KUBE-SERVICES \
        -d 34.68.37.70/32 \
        -p tcp \
        -m comment --comment "ci/ci-web:tsa loadbalancer IP" \
        -m tcp --dport 2222 \
        -j KUBE-FW-3WZFA3OBZWICIHEP# in this "catch anyone that landed here" rule, we jump to the "mark to drop"
#
-A KUBE-FW-3WZFA3OBZWICIHEP \
        -m comment --comment "ci/ci-web:tsa loadbalancer IP" \
        -j KUBE-MARK-DROP# mark drop just puts a mark so that a "catch all with this mark" rule can than
# act on it.
#
-A KUBE-MARK-DROP \
        -j MARK --set-xmark 0x8000/0x8000# acting on those marked as `0x8000`: drop!
#
-A KUBE-FIREWALL \
        -m comment --comment "kubernetes firewall for dropping marked packets" \
        -m mark --mark 0x8000/0x8000 \
        -j DROP

^ (from the host)

[cirocosta]

Thus, the final condition is to:

• allow any "dns resolution" (actually, it's more of "any traffic to TCP/UDP:53)
  • (perhaps, we should restrict this to only external dns'es? if so, we'd need to be able to provide custom dns settings to the Concourse chart)
• allow egress to ci-web pod in the ci namespace
• allow any out except to 10.0.0.0/8 (internal nets)