battery: define a source-grounded state-of-charge model before wiring status

[forkwright]

Finding

Phase-07 planning and the compiled-only BatteryMonitor use conflicting voltage-to-percentage curves: the plan states 3.7 V = 50% and 3.4 V = 10%, while source maps them to roughly 30% and 5%. Neither is tied to a qualified cell, load, temperature, PMIC ADC source, hysteresis, or uncertainty model.

Evidence

• crates/thumos/src/battery.rs: the hard-coded piecewise curve maps 3700 mV and 3400 mV differently from Phase-07 REQ-13.
• docs/capability-inventory.toml: BatteryMonitor is compiled-only and status-bar battery is still a placeholder.
• Phase-07 decisions correctly say calibration belongs on the M7, contradicting its own fixed generic points.

Why this matters

A fabricated percentage can trigger premature shutdown, conceal low charge, mislead field use, and invalidate power profiling. Hardware measurement should calibrate a defined software model, not choose silently between contradictory constants.

Done when

• Identify the stock/replacement cell and source-ground the MT6357 ADC/charger/fuel-gauge inputs, units, scaling, error/status bits, and read schedule.
• Define a versioned SoC model with load/temperature/charge-state handling, smoothing/hysteresis, uncertainty, stale/unavailable behavior, and safe low-voltage policy.
• Put all constants behind one configuration/model owner; UI reports unknown rather than a guessed 0% or stale value.
• Host tests cover boundary/noise, charge/discharge, stale/error, calibration migration, and shutdown thresholds.
• After software/source acceptance and kernel: tracking anchor for the compiled-but-unwired surface #753 integration, an operator-owned M7 session may calibrate/validate the curve.