Retry failed upgrade downloads, reusing the previously downloaded data, and fallback to the DNS mirror pool

This is the solution called "C + fallback DNS pool" on #15875.

Incidentally, since the SocksPort we're using has IsolateDestAddr and
IsolateDestPort, this also gives us "C + new Tor circuit" for the last retry in
most cases (that is, unless the mirror picked from mirrors.json, that we tried
to download from previously, is in the DNS pool, and we pick it again from from
the DNS pool).

This should allow the Upgrader to recover, transparently for the user, from
a number of failure modes, such as:

 - We picked a mirror that's down or flaky.
 - We picked an out-of-sync mirror that lacks the data we need to download.
 - The user's Internet connection is flaky.
 - We picked a flaky Tor circuit.

Regarding the added test scenarios:

 - For the "Successfully resuming an interrupted download, from the same mirror"
   scenario, I failed to find a simple enough way to set up a webserver that would
   fail once and then start working again, so I implemented the webserver mocking
   code directly in Tails::IUK::TargetFile::Download, which is not ideal.

 - For the "Successfully resuming an interrupted download, using the fallback
   mirror pool" scenario, I could have run a broken webserver and a functioning
   fallback one. But once I had the mocking code mentioned above, it was vastly
   simpler to just use it here as well.

refs #15875