Implement advertisement and popover identification based on DOM sample data

[vlofgren]

Continuing PR #201, this change set adds a new internal API for fetching DOM samples for a given domain, which gets called from the converter. The converter analyzes the DOM sample, and adds new document features, which are acted upon by the index to better penalize documents with advertisement.

--

The change also largely retires the quality property on documents, which was used to filter out documents with too much javascript. This is a very blunt instrument that threw out a lot of perfectly valid search results. Using DOM sampling to assess what sort of scripts is a better way of doing things.

Additionally, the converter is less strict on filtering out documents based on length. Instead, short documents are assigned a feature bit so that they can be ranked lower.

--

In order to easier adjust the request classification model, a GUI was built for viewing the sampled traffic. Since the data is interesting, this was made public and accessible under the site inspector tool.

• Implement API + Client
• Implement classifier
• Integrate classifier data into converter
• Retire old quality measure when DOM sample data is available
• Integrate new features into index
• Verify UI filters act properly
• Build inspection UI
• Tune search result quality in test environment
• (fix misidentification of blogspot and wordpress sites as having ads)
• Fix UI scaling for mobile
• Verify light mode layout doesn't look like crap
• Ensure system doesn't break for non-prod deploys that don't have the domsample api provided