Bug: the junction-temperature warning reports a part's rated maximum, not what the board dissipates #274

What were you trying to achieve?

Read W-SYNTH-THERMAL-001 on the Class-A boards as a statement about those
boards, and act on it.

What are the expected results?

A junction-temperature estimate for a part on this board: the power it
actually dissipates here, through its thermal resistance.

What are the received results?

The part's own maximum, on every board that uses it, whatever the load:

W-SYNTH-THERMAL-001 | U3 reaches about 107 C (1360 mW over 60 C/W)
W-SYNTH-THERMAL-001 | U1 reaches about 107 C (1360 mW over 60 C/W)

The first is protected_power_entry, where the AMS1117 feeds an indicator LED
and a header. The second is i2c_sensor_node, where it feeds an STM32G030, a
BME680 and a 24LC256. The two boards differ by an order of magnitude in load
and report the same temperature, because 1360 mW is
ams1117_3v3's declared max_power_mw rather than anything either board
asks of it.

On the sensor node the real figure is 5 V to 3.3 V across a budget of about
300 mA — the sum of the three devices' declared max_current_ma — so 510 mW,
31 C of rise, about 56 C. That is a different conversation from 107 C.

Steps To Reproduce

1. Put an ams1117_3v3 on a board with one LED on its output.
2. synth validate.
3. W-SYNTH-THERMAL-001 reports 107 C.

In what environment did you encounter the issue?

synth from main at dddda6f.

Additional information you deem important

The load is already computable: E-SYNTH-POWER-010 sums the declared
max_current_ma of every device on a regulator's output net to decide whether
the regulator is over budget. The same sum, times the input-to-output drop,
is the dissipation — and the input voltage is available wherever the board
declares its rails with power "NAME" <V>.

The honest fallback when a rail is not declared is to say so, rather than to
substitute the part's rating: a warning that fires identically on every board
using a given part carries no information about any of them, and the first
thing a reviewer does with it is learn to ignore it.

Found while making the Class-A boards standalone: #268, #269.