fix(pixii): preserve measured AC power sign

[segran2]

Fixes #154.

Problem

Pixii register 40083 (measured AC power) already follows FTW site signs on the reported PowerShaper hardware: charging is positive and discharging is negative. The driver inverted that value as if it used the same generator-frame convention as the setpoint. During discharge, for example, -1231 W became +1231 W and Core could report power_wrong_direction.

Change

• preserve the sign of measured AC power for control_power_w
• leave setpoint/command sign conversion unchanged
• add regression coverage for both negative discharge and positive charge feedback
• bump Pixii from 2.1.7 to 2.1.8 and refresh catalog metadata
• update the changelog

Evidence

Issue #154 contains PowerShaper site telemetry showing battery_ac_w = -1231 W while the battery was physically discharging. That matches the repository contract: battery discharge is negative.

Validation

The existing Pixii status test now asserts both directions:

• -1400 W measured AC -> -1400 W control_power_w
• +1400 W measured AC -> +1400 W control_power_w

GitHub CI should run the repository checks for this PR. No command, heartbeat, calibration or safety-policy behavior is changed.

Paired Core update: srcfl/ftw#1504. Core bundles Pixii 2.1.8 and tests measured power and command feedback in both directions.

[segran2]

Hardware validation

Tested this PR on a real Pixii Home system using the Pixii 2.1.8 driver from commit 1d3babb02d81f2ba3e47cdc864a5b1ff1b1a0818 as an FTW user-driver.

Both power directions were verified through FTW Core's measured control_feedback.

Charging

requested_w  +5625.58 W
sent_w       +5625.58 W
readback_w   +5626 W
actual_w     +5661 W
evidence     measured
readings_fresh true

The driver status at the same time also reported positive battery power (bat_w ≈ +5154 W).

Discharging

requested_w  -1504.74 W
sent_w       -1504.74 W
readback_w   -1505 W
actual_w     -1501 W
status       following
reason       power_observed
evidence     measured
readings_fresh true

The driver status at the same time reported bat_w ≈ -1598 W.

This confirms on hardware that measured command feedback now preserves FTW site signs in both directions: charging is positive and discharging is negative. Command/setpoint conversion also remains correct. No power_wrong_direction was observed during these tests.

[segran2]

Origin of the regression

I traced the sign regression to commit 424f1e5c75e4cf4d9ada3ca016c6cba5c428e893, merged via #149 on 2026-10-01.

That change introduced the assumption that Pixii AC power register 40083 uses the same SunSpec generator-frame sign convention as the setpoint, and therefore negated it for control_power_w. The commit message explicitly noted that the sign had not yet been hardware-verified.

Hardware validation on a Pixii PowerShaper now shows that register 40083 already uses FTW site signs:

• charging: positive
• discharging: negative

Observed discharge validation:

• requested: -1504.7 W
• readback: -1505 W
• measured actual: -1501 W
• FTW control status: following / power_observed

Charging was also validated with positive measured power. This confirms that control_power_w must preserve ac_w rather than negate it.

The originating commit is authored/signed off by Fredrik Ahlgren and also credits Claude Opus 5.5 as co-author.