cpu_seq: fix set_power_state idempotency for substates

[hawkw]

PR #2064 was supposed to change the Gimlet and Cosmo CPU sequencer API so that requesting an idempotent power state transition (i.e. A0 -> A0) returns Ok(Transition::Unchanged) rather than
Err(SeqError::IllegalTransition). Unfortunately, it doesn't always do that in some cases.

This is because the match arm in which we detect no-op power state transitions tests that the current and requested PowerStates are equal. But, unfortunately, the PowerState enum has more variants than just "A2" and "A0" --- it also represents substates of A0 and A2, such as A0PlusHP and A2PlusFans.

In particular, when a compute sled is fully up and running, it's actually not in PowerState::A0, it's in PowerState::A0PlusHP, because the NIC hotplug controller is enabled. The PowerState::A0 and PowerState::A0PlusHP enum variants are not equal, so when a sled that's actually in A0PlusHP is told to go to A0, it doesn't match and falls through to the error case incorrectly.

I'm pretty sure the reason we didn't realize this earlier is because, while I did test both A2->A2 and A0->A0 transitions when I was testing PR #2064, I would send the A0->A0 request more or less as soon as the system reached A0. I didn't wait for the host OS to come up before testing it, so the system was still in PowerState::A0 and not PowerState::A0PlusHP. Whoopsie.

This commit makes this operation idempotent in those cases by treating A0PlusHP->A0 and A2PlusFans->A2 transitions as idempotent successes rather than IllegalTransitions. I did not change the behavior in A0 substates that indicate a CPU reset condition (A0Reset and A0Thermtrip), as in those cases, we require an explicit transition back to A2 before the system will return to A0.

Fixes #2271