fix(scale): harden HDS sleeping reconnect follow-up

[ODevStudio]

Summary

Follow-up to merged PR #867, based on review against the current decentespresso/openscale HDS firmware behavior.

This draft addresses two lifecycle gaps:

1. Display-off reconnect state — OpenScale intentionally restores the display after a BLE disconnect unless it is in true SoftSleep. A Decaid reconnect while displayOff was therefore able to leave the physical display on while Decaid still considered the scale asleep. Sleeping reconnects now re-assert the shared 0A 00 display-off command.
2. SoftSleep wake confirmation — OpenScale queues the 0A 04 00 wake from the BLE callback and applies it later on the main loop. A successful GATT write therefore proves that wake was queued, not that it was applied. Decaid now keeps the SoftSleep-exit obligation until a post-wake status frame is observed; the existing bounded exit retry is reused when confirmation does not arrive.

The sleeping lifecycle also retains the last positively negotiated power-off/SoftSleep capability evidence across the reconnect boundary. The ordinary per-connection profile still resets to conservative and is re-negotiated on wake, so this does not relax fresh-connection negotiation for normal awake connections.

OpenScale reference

Current OpenScale behavior used for this follow-up:

• BLE disconnect calls restoreDisplayAfterBleDisconnect(), which restores the display unless b_softSleep is true.
• BLE SoftSleep enter/exit commands only queue main-loop work.
• displayOn() queues a BLE status response, and main-loop pending commands are applied before the queued BLE status response is emitted.

Tests

Added focused regressions for:

• display-off reconnect re-asserting darkness;
• retained HDS power-off evidence surviving a sleeping reconnect;
• wakeDisplay() remaining pending until a post-wake HDS status frame arrives.

Verification

• Reproduced all three CI failures in the shared display-off reconnect test helper, then asserted exactly one display-off packet rather than no writes.
• Focused disconnect, OpenScale follow-up, and lifecycle suites: 41 tests passed.
• flutter analyze --no-pub: no issues found.
• flutter test --no-pub --concurrency=1: 4255 passed, 1 skipped on Windows. Used the pinned plugin assets, QuickJS DLL and OpenSSL on PATH; serial execution avoids local port contention between web UI suites.
• Changed test formatting and git diff --check: clean.
• No additional hardware pass has been performed for this follow-up. GitHub CI will rerun after commit 539caafd.

Refs #867.

Linked Issue

N/A - maintainer follow-up to merged PR #867.

Impact

Sleeping HDS reconnects reassert display-off, and SoftSleep wake waits for status confirmation. No API/spec, database migration, skin, or plugin contract changes. The CI follow-up only updates the existing reconnect test helper to assert the exact display-off packet instead of expecting no writes.

Contributor Responsibility

• I have reviewed and understand all changes in this PR and take responsibility for their correctness, security, behavior, licensing, and provenance, including any AI-assisted or AI-generated work.

[tadelv]

Reviewed adversarially against AGENTS.md, the surrounding Decent Scale lifecycle code/tests, and OpenScale's disconnect behavior. The underlying issue is real: OpenScale can restore the display after BLE disconnect unless soft sleep is active, so retaining the sleep-time capability/profile state and reasserting Display Off on reconnect is appropriate. SoftSleep is correctly not reissued, normal data-channel/profile negotiation resumes on wake, and the added regression coverage exercises the relevant reconnect/wake paths. I did not find a merge-blocking correctness or architecture issue in this PR.