This adds a special file, "pubkeys.gpg", to test ACI server's fileset.
If it is available, the server will send the ac-discovery-pubkeys meta
tag too.

Asking the user whether the user really trusts the key can happen in
two situation - when running rkt trust or when running rkt fetch with
--trust-keys-from-https set to false. Split out the function replying
to the question from the function calling rkt trust. The new function
will be used when testing various trust scenarios during rkt fetch.

Also, changed the function a bit to take a boolean parameter telling
whether to answer yes or no to the question.

If the signature file already existed, gpg asks us to confirm that we
want to replace the old signature with the new one. We don't handle
the case, so we hang until the tests time out.

Fix it by deleting the signature file before creating a new one.

So we can pass a gpgkey instance instead of just an index.

This is to allow the signature file be called differently than image
filename + ".asc" suffix. Will be useful for generating wrong
signatures to be uploaded to test server to impersonate good ones.

Just a single place doing it, instead of getting the value of the
FUNCTIONAL_TMP environment variable.

The first two keys will be used as good and wrong key respectively in
the trust discovery test. But we will need a separate key for singing
stage1 image, so that the key does not interfere in the tests. Signing
the stage1 image is necessary since recently for stage1 images passed
to rkt via --stage1-path.

This basically tests all the combinations of signatures (missing,
signed wrong image with good key, signed wrong image with a wrong key,
good), keys (both to be trusted and on the server; missing, wrong or
good), trusting keys from https (rejecting the review, accepting the
review and actually trusting them).

It doesn't test the --signature parameter (which tells rkt to use some
local signature instead of the remote one), it probably could be done
in a separate test, to avoid complicating the test even more.

It is using stub stage1 image to get the 4-5x speedup, so even if
there are 108 combinations, it still should take less than a minute.