When getting a resource's fail count, the scheduler checks whether failures are expired. As part of that, it checks whether the resource has a failed action with `on-fail="block"`, in which case the failure never expires.
Currently, this uses an XPath to search the configuration for `on-fail="block"`. However, that does not detect `on-fail="block"` when used via id-ref, operation-specific meta-attributes, or op_defaults, and does not evaluate rules. Also, it determines failure by simply comparing the actual exit status against the expected exit status, which does not account for all the situations handled by `remap_operation()`.
Instead, we should (temporarily) instantiate an action, which considers everything when calculating on-fail, and remap exit status appropriately.