[gateway] Report CPU Tctl as a dimensionless metric

[hawkw]

Currently, the AMD SP3 host CPU's T_ctl value is reported to
Oximeter as an instance of the hardware_component:temperature metric
with sensor = "CPU". This metric's unit is in degrees Celcius.

This is incorrect. T_ctl is not a physical measurement from a
temperature sensor, but an internal parameter of the CPU's thermal
control loop in synthetic dimensionless units that range from 0-100. For
details, refer to this comment and oxidecomputer/stlouis#5.

This branch adds a new hardware_component:cpu_tctl metric. Unlike the
hardware_component:temperature metric, this metric has no unit, as it
is a dimensionless value. The SP sensor metrics task in MGS has been
changed to special-case temperature measurements where the sensor name
is "CPU" and the device kind is "sbtsi" so that they are reported using
the hardware_component:cpu_tctl metric rather than the
hardware_component:temperature metric.

In the future, I think a more ideologically correct solution to this
would be to add a variant to the
gateway_messages::measurement::MeasurementKind enum to represent
dimensionless measurements (or, perhaps, specifically for
T_ctl?), and change Hubris to report the SB-TSI
T_ctl using that instead. However, this looks like it would
probably be a somewhat more complex change in Hubris, as the thermal
control loop would still need to consider that measurement. This change
introduces the new metric type and fixes the problem of us reporting
incorrectly-labeled metrics, so I think it's worth handling it this way
and then going back and making the Hubris change later.

Fixes #6634

[hawkw]

Agh --- this test failure is because i forgot to add the simulated component to the MGS tests. Will fix that.

[bnaecker]

This is a solid, very practical fix, thanks! I agree that we can look for a better long-term solution, but this is definitely good enough for now. Just one small comment suggestion, but LGTM!

[hawkw]

@bnaecker Thanks for the speedy review! I belatedly realized that we should also distinguish between T_ctl and other temperature metrics for the purpose of missing samples and error metrics, so 6552ab0 fixes that. I don't know if you feel the need to review again or if you're still good with the previous review.