Add instruction mechanism to run instruction even if prior instruction failed

coin today maintains a list of scheduledInstructions that are run regardless of whether the main instructions fail. But this list is currently hard-coded (to upload core dumps, print sscache stats, etc).

There is currently no generic way to mark an instruction as always running.

A convenient way to solve this would be to add a `when` attribute to instructions, that defaults to `PreviousInstructionPassed`, but can be changed to `PreviousInstructionFailed` or `Always`

If this attribute/property is added to Group instruction, all the instructions in that group should inherit the behavior of the group.

We could consider disable_if/enable_if instead of a separate when property, but I'm not sure if those are enabled as the job is run.

enable_if:
    condition: previousInstruction
    equals_value: Failed

Thoughts?