Having self.diag and self.t.diag was confusing as one referred to TAP
output (self.diag) while the other referred to "task diagnostics"
self.t.diag.
self.diag is now gone and was replaced by self.tap with the same
behavior.
* When testing multiple clients vs server versions, repeating client
tests is unnecessary. By setting the env variables TASKW_SKIP and
TASKD_SKIP it will now be possible to skip all task (client only) and
taskd (client + server) tests, respectively.
