Before applying this patch, `cluster/kubectl.sh` failed to find
`kubectl` binary in x86 environment.

e.g. The latest release v0.17.1 contains following binaries:
https://github.com/GoogleCloudPlatform/kubernetes/releases/tag/v0.17.1

```
$ tree platforms/
platforms/
├── darwin
│   ├── 386
│   │   └── kubectl
│   └── amd64
│       └── kubectl
├── linux
│   ├── 386
│   │   └── kubectl
│   ├── amd64
│   │   └── kubectl
│   └── arm
│       └── kubectl
└── windows
    └── amd64
            └── kubectl.exe
```

Ignore /proxy from performance metrics

Fix WARNING when creating firewall during e2e tests

Extended wait for cluster size timeout.

Extended wait for cluster size timeout in resize nodes e2e test to fix #8715.

Description of ImagePullPolicy wrong in v1(beta3)

update docs/volumes.md to v1beta3

WIP: Continue to support (deprecated) publicIPs

Not doing so breaks e2e tests and people that may be using them,
even though we will eventually want to stop supporting this now
that we have better alternatives for typical use cases (NodePort)

Nodeport GCE e2e fixes

Re-add the cluster-name flag to ControllerManager

WIP: ServiceType & NodePort work

Although it takes some time to destroy a load balancer, we hide this complexity
from the user.

We still don't actually enable this though!

Including some refactoring of IP allocator

We need to keep them in the API so that we can round-trip between versions.

We prevent it from being set by validation

Better logging for e2e start (debug failing pods)

Use e2e framework in more tests

Fix kubelet tests for systems without functional dns to look up localhost

This will replace publicIPs

Don't clobber GINKGO_TEST_ARGS in hack/jenkins/e2e.go

Start e2e framework; print events