losing CPU temperature induces a fan party

[nathanaelhuffman]

I'm un-convinced that this is good behavior, loss of the CPU temperature does potentially indicate a system wide problem and does make a thermal control loop hard, but I'd propose that a more correct behavior here would be to log the fact that this happened, optionally spins fans up (incrementally) to provide margin if the system was hot, put something in both an ereport and a ring buffer and desequence.

As it is, there's no up-stack reporting that something is wrong and we're left hoping someone notices the fans totally spun up and reports a problem, we should change this to provide automated failure reporting and make sure we can find out that something happened in a reasonable way.

[hawkw]

I added the "ereport" label to this since we definitely want to be reporting an ereport in the case of losing CPU temperature. I do think that's somewhat distinct from whether it's correct for the thermal control loop to go to 100% PWM in this case or not.