For example, 'apply <env' currently operates as a no-op. With the
introduction of simple environments in PR #131, 'apply <env>' should
perform basic validation such that:

1. The user has added the environment that is being deployed against to
their Ksonnet project.

2. The URI in the environment's spec file that the user wishes to deploy
to should correspond to at least one cluster location as listed in
kubeconfig.

If either of those conditions are not satisfied, the kubecfg user will
receive the corresponding error.

In addition, this commit will set the kubectl --cluster flag to point at
the cluster listed by the environment URI.

Currently, commands take either an <env> or a '-f' argument but not
both. With this commit, we are allowing both args to be provided. The
behavior is expand the files passed by the '-f' flag and deploy the
objects to <env>.

This commit is a follow-up to the discussion in the ksonnet.next design
doc, in which users consistently expressed their preference that the
`update` command be called `apply`.

NOTE: We have renamed `pkg/kubecfg/update_test.go` ->
`pkg/kubecfg/apply_test.go`, but we copy the `update`'s integration test
file. The reason is that the `update` unit tests actually test things
like GC (rather than the command itself), so the file name is
inconsequential. On the other hand, the integration tests test the
commands themselves, so it is important to have two copies, one for
`update` and one for `apply`.

Currently, if the user wants to deploy a ksonnet application, and that
application uses some vendored library, or something in the `lib/`
directory, they will have to pass the appropriate `-J` flags into the
command themselves.

This commit will automatically add these whenever we're in an app
directory and a command is issued. In particular, even if the user
passes in the `-f` flag (rather than an environment name), we'll still
add library paths to the command if we're in a ksonnet app directory.
This is meant to capture the case that a user wants to update one
resource in particular in a ksonnet application.

If the current kubeconfig context is declared with an
explicit (non-empty) namespace, then `mergedKubeConfig.Namespace()`
returns that *config file* namespace value, ignoring an explicit
`--namespace` command line flag.

See https://github.com/kubernetes/client-go/issues/288 for upstream
client-go bug/discussion.

This change works around the issue by just checking the Namespace
override ourselves before falling through to the regular function.

Further fixes #103

This reverts commit 020d5da0.

The change broke --namespace (and other flag) handling, because it
recreates the `clientConfig` global singleton multiple times - only
some of which are correctly tied to the relevant flags.

This commit will separate the application level logic for the 'show', 'delete',
'validate', and 'diff' commands from the Cobra logic in the cmd/ package.  The
application level logic will be placed in pkg/kubecfg/.

The ksonnet.next design doc specifies the core kubecfg verbs (i.e., the
ones listed above) to all have the form:

  kubecfg <verb> [<env-name>|-f <file-or-dir>]

That is to say, each of these should be able to take either an
environment name, or a `-f` flag with a list of files and directories to
apply `verb` on. In the case of the environment, we will apply `verb` to
every component in the `components/` directory.

This commit implements this behavior for all these verbs.

This commit will introduce the `template.Expander` abstraction, which is
meant to abstract over an invocation of the Jsonnet VM. Specifically, it
provides facilities for users to provide (e.g.) Jpaths, ext vars, and so
on.

The main justification for this change is:

* We need a common way for the `pkg` and `cmd` packages to interact with
  the Jsonnet VM.
* The `utils` package is already too much of a catch-all.
* It's easier to think about an invocation of the Jsonnet VM when we
  additionally encapsulate the parameters we pass to it on every
  invocation in kubecfg.

This will make it possible to inject fake dependencies, once the fake
dynamic client makes it into a client-go release.

(other commands will be moved similarly in future PRs)

- Move fqname to utils
- Add ResourceNameFor and use it to get a lowercased/pluralised
  version of the resource name

Fixes #15

Improve interactive output experience by switching to a logging
library (logrus) that allows customising the output format.

This also removes the glog command line flags.  Replaced with a
simpler `--verbose`/`-v` option:
- quiet by default (warnings and errors only)
- `-v` for progress info
- `-vv` for debug

Fixes #34

Fixes #35

Uses github.com/yudai/gojsondiff for the heavy lifting.

The goal is to make a best-effort attempt at reducing the number of
"crash-restart" loops required to bring up a group of interdependent
resources.

The current implementation is very simple and just sorts into 3 tiers:
- Namespace, ThirdPartyResource, StorageClass
- everything else
- Pods or similar (Pod/Job/Deployment/DaemonSet/StatefulSet)

- Support the standard client-go/kubectl flags
- Switch object types to use client-go rather than apimachinery
- Implement update subcommand