thermal-service: Detect potential invalid/erroneous sensor readings

Currently the thermal service will handle complete sensor failure adequately (as in, failures to communicate with the sensor at all) but should implement a way to detect if readings might be invalid even if we can communicate with the sensor. For example, if the sensor is repeatedly reporting temperatures outside of an expected range we should treat these as suspect and mark the sensor invalid and take appropriate actions (such as possibly telling fans that depend on the sensor to run at max speed).

Would also want to detect momentary spikes and discard them. For example if sensor has been reading steadily within a certain range and then suddenly we receive a sample that is considerably higher which may trigger a PROCHOT, wait to see if we get that sample several more times before actually triggering a prochot.