The intent of the new test case is to catch general breakages in
the fsck_tag() function, not so much to test it extensively, trying to
strike the proper balance between thoroughness and speed.
While it *would* have been nice to test the code path where fsck_object()
encounters an invalid tag object, this is not possible using git fsck: tag
objects are parsed already before fsck'ing (and the parser already fails
upon such objects).
Even worse: we would not even be able write out invalid tag objects
because git hash-object parses those objects, too, unless we resorted to
really ugly hacks such as using something like this in the unit tests
(essentially depending on Perl *and* Compress::Zlib):