Counting failing xfail tests as ok

[lorcap]

Currently, if a test marked in Pytest as xfail (expected to fail) actually fails, it increments the count of not-ok tests. My proposal consists in adding a new option --tap-xfail-ok which counts non-strict failing xfail tests as ok.

Rationale: an xfail test is neither a fail nor a pass test. User might want to run xfail tests and successfully pass the CI, knowing that those tests will be fixed in a future version.