A deal with a bogus signature would still get stored (before producing
an error). That would prevent a properly signed deal coming later from
being stored. This fixes the bug.

The test was also written in a flawed way. It measured two failure
situations ("wrong index" and "wrong deal") after a success situation. But
written that far down, those measurements could well end up measuring
failures due to a duplicate deal instead. This changeset also patches
the test flaw.