power: add charge-aware auto sleep mode for Pocket EVO

[jesherman]

Summary

Adds an "Auto adjust" sleep-mode option to armada-control that resolves per suspend: s2idle when a charger is present, deep sleep otherwise.

Why

Deep sleep powers down the ADSP on the Pocket EVO, which kills the PD contract and the battery-manager firmware:

• a plugged-in EVO in deep sleep drains instead of charging (0 W in, ~1.5 A out)
• the UCSI firmware can come back in source mode on wake (2 of 3 wakes in testing), requiring an ADSP restart to recover
• charging is uninterrupted in s2idle (ADSP stays alive; ~15–20 W through sleep, clean wake, charging resumes immediately)

"Auto adjust" picks per-suspend so the charge state can't be misconfigured: plugged → s2idle (charging continues), unplugged → deep (battery savings).

Design

The choice is resolved at suspend time, not stored as a kernel mode:

• device-env — accepts auto in /etc/armada/sleep.conf and resolves it to s2idle/deep from the qcom-battmgr USB supply (online=1 → s2idle, else deep). Every existing consumer (suspend-dispatch, device-quirks) sees only a concrete mode — no change to the suspend path.
• armada-control — advertises the new option in available_sleep_modes(); select_mem_sleep() deliberately skips writing /sys/power/mem_sleep for it (it is a resolution layer, not a kernel mode).
• decky system.py — adds the "Auto adjust" entry to the sleep-mode menu.

No kernel changes; no behavior change when the option is not selected (existing modes unchanged).

Verification (on-device, hybrid 7.2.0-rc7)

• Plugged s2idle: ~15–20 W held through sleep, PD contract survives, battery charging on wake, policy re-establishes 28 W in ~90 s
• Unplugged deep: no charging (battery preserved)
• Charger attached during deep sleep: clean wake, PD renegotiates to PPS, charging resumes — no wedge
• 11-check automated harness (resolution, parsing, explicit-mode precedence) all passing

Closes the sleep-charging gap for Pocket EVO owners without forcing a single sleep mode on everyone.