Splash Generation Improvements

[leichter]

• Reuse Phantom processes in splash generation. It would be nice to reuse phantomjs processes when generating splash images (and just open a new page) and not have to create/kill the processes each time. The tricky thing here would be ensuring that one splash generation is finished before telling the phantom process to start on another. Could almost turn this into a task queue of sorts.
• Download thumbnail images locally. Rather than templating in the urls we receive directly from the client, download the thumbnail images and save them locally. This way we could avoid doing hacky logic to switch protocols, and we could take advantage of caching the thumbnails for multiple requests.

[leichter]

Some version of the first of these two bullet points needs to be implemented for the purposes of throttling requests to the splash generation endpoint. I was able to basically shut down a test ec2 instance by asking it to generate 100 splash images at once.