Make the Google Drive CDN work properly

[jikamens]

The Google Drive CDN wasn't working properly for a number of different
reasons. This commit contains many fixes to make it work. In
particular:

• The URL format that was being generated for files in Google Drive
  (https://random-stuff.googledrive.com/host/random-stuff/file-path)
  no longer works. The code in this commit uses the URL format
  https://drive.google.com/uc?id=_file-id_, which unfortunately
  requires a redirect each time the file is fetched, but a fetch with
  a redirect is better than the CDN not working at all.

• The code for uploading files to Google Drive was assuming that files
  in the same chunk with the same title were the same file. This was
  causing all sorts of problems, such as attempting to delete files
  multiple times, attempting to access files that had been deleted,
  uploading files that had already been uploaded, comparing local
  files with the wrong uploaded files to determine whether they needed
  to be uploaded again, etc.

  I've fixed this by adding properties to the uploaded files
  indicating their full path, and using the properties to disambiguate
  files with the same title. We also use these properties to allow us
  to locate uploaded files and repopulate the ID catch (described
  below) after it is purged.

• Because everything before "file-path" in the old-style URLs was
  the same for all URLs, it was inexpensive and fast to generate
  Google Drive CDN URLs. Now, however, every URL has to have the
  Google Drive file ID in it. We can't afford to query Google Drive to
  determine all of the file IDs every time we're rendering a page.
  Therefore, I've added a database table to cache the mapping from
  remote file path to file ID. I've also added a command to the
  Performance menu in the title bar to purge the contents of this
  table in case it somehow gets corrupted.

• I've changed the code that returns errors to the user when an upload
  to Google Drive fails, so that instead of just reporting the
  "reason" from the Google Drive error, which isn't sufficiently
  detailed, it sends back all of the errors that Google Drive sent,
  JSON-encoded so they're relatively easy to read and figure out
  what's going on.

• CSS files with relative URLs in them were getting minified and their
  URLs were getting turned absolute, but the scheme and domain name
  were not being prepended to the beginning of these URLs. As a
  result, once these URLs were minified and uploaded to the CDN, these
  URLs could no longer be loaded, since the origin was Google Drive
  rather than the blog host where these URLs actually reside. I've
  fixed this by adding a get_domains function to the CDN which
  causes URLs to be fully qualified in minified CSS files.

  Note: Unfortunately, these URLs are not being written to come from
  the CDN, even if the relevant files are present there. I spent a
  good long time trying to figure out how to make this work and
  couldn't; there doesn't seem to be a concept in the code of doing
  URL replacement inside CSS files. This is no worse than it was
  before, but it's something it would be nice if we figured out how to
  make better.

side note: see #523