test-lib: test_expect_failure() expected failure is ok

test-lib: test_expect_failure() expected failure is ok

The Git test-lib test_expect_failure() behavior is to call separate
test_known_ok/test_known_broken_failure helper tags which puts the
"known failure" results into separate buckets. This is the opposite
of what I want. I want test_expect_failure() to be the true-opposite
of test_expect_success(): if we expect a test to fail (i.e. exit 1),
then say the test passed if the command failed.

* tests/test-lib-functions.sh (test_expect_failure): Change
  test_expect_failure() to be the true-opposite of
  test_expect_success(): test succeeds if command exits with error
  code.