Admin server with DELETE /api/pool handler

[parente]

Use case: I want to docker pull or docker build a new container image on the tmpnb box. I don't want containers already in the pool based on the old image to linger indefinitely until they are used up. I want to immediately drain the pool of idle, available containers and replenish the pool with containers from the new image. While doing so, I do not want to disturb containers that are currently in-use. In practice, I should be able to do something as simple as:

eval `docker-machine env my-tmpnb-host`
docker build -t my-notebook-image .
docker exec -i tmpnb-pool curl -X "DELETE" http://127.0.0.1:10000/api/pool

to get the desired effect.

Since pool draining is a costly operation, this PR adds a new "admin" server that listens on a port separate from the main tmpnb server. The admin server only listens on localhost by default and supports an optional ADMIN_AUTH_TOKEN env var as a shared-secret client authorization mechanism.

The admin server currently implements a single handler for DELETE /api/pool that implements the behavior described above. However, it can grow to support other admin operations in the future (e.g., PUT to change config settings like total number containers in the pool).

I'm open to other verbs and resource names. DELETE /api/pool was the first thing that came to mind.

[captainsafia]

Neat-o. You might want to add a note about configuring the ADMIN_AUTH_TOKEN env var to the qucikstart section in the docs, too. Code comments, inline.

[rgbkrk]

Awesome, thanks for doing this @parente!

[rgbkrk]

/api/pool is fine. There's only one pool with a single tmpnb node, so it's the resource.

[parente]

I'll push one more commit later today with the README additions. Then it should be GTG. Thanks for the quick reviews!

[parente]

Updated the README.

[captainsafia]

Looks good to me.

I will leave merging to our grown up-in-residence, @rgbkrk.

[rgbkrk]

Ha, you can merge too @captainsafia

[parente]

DH build is failing. So either I broke it by de-duping the requirements install step in the Dockerfile or DH is just having issues.

Running make tmpnb-image again locally works fine so I'm guessing it's the latter. Will keep an eye on it.

[parente]

After 3 failed attempts with no useful info in the log as to the failure, I tickled the Dockerfile a bit based on the hunch that maybe the earlier version of Docker used on DH for builds was getting tripped up by the ADD to a path without a trailing slash. Switched to using the safer COPY directive and made the path more explicit in 8a35b355147afd76d52ba601d22cb4245a7d07cb. Build now succeeds.