JBD: support reliable MOS control and extended diagnostics

[tednv]

Summary

• use the complete JBD factory-mode transaction for charge/discharge MOS commands
• reconnect when keep_alive is disabled, preserve the other MOS state, ignore already-satisfied retained commands, and verify each write
• expose read-only JBD identity, firmware, capacity, protection, balance, voltage-curve, and configuration registers through MQTT discovery
• support display_name and comma-separated temperature_names without changing MQTT topics or entity unique IDs

On supported JBD firmware, this can re-enable charging when an undervoltage event has left the charge MOS disabled, including states where standard app controls are unavailable. Active protection conditions are not bypassed.

Scope

Extended JBD settings are read-only. Existing charge and discharge switches remain the only settings this change writes.

Validation

• tested through Home Assistant with a JBD 14S BMS running firmware version 3.2
• verified the MOS-control sequence and state readback against the physical BMS
• 23 JBD tests passed
• 5 MQTT discovery and sampling tests passed
• Python compilation and git diff --check passed

Closes #418

[fl4p]

Thanks, merged and released in 2.25. I changed a few things on top (63c3df7, e46a566, ef589fe):

• Settings read leaves factory mode with 00 00. 28 28 also saves to EEPROM (dbus-serialbattery lltjbd.py), and that ran on every add-on start. Switch writes still exit with 28 28, as you tested.
• Settings registers time out after 2 s, and the read stops after 3 unanswered in a row. At 16 s each, a clone could sit in factory mode for ~15 min before the first sample.
• Register 0x16 is no longer the HA device identifier. Clones share factory serials and would merge into one device.
• display_name / temperature_names work for every BMS type, not only jbd.
• Switch commands from MQTT run concurrently with the sampler. Its disconnect could cut the sequence between 56 78 and the exit, leaving the BMS in factory mode. Register I/O, disconnect and the periodic reconnect now share a lock. With keep_alive: false, a command is applied at the next poll instead of opening its own connection.
• A refused write no longer drops the sample when the algorithm made it. Under an active protection, the same state isn't rewritten for 10 min (each write is an EEPROM save).
• A command a protection is forcing is still written once, so the FET stays there after the protection clears. The other FET keeps its last commanded state, not the one a protection set.
• A status != 0 reply raises right away instead of a 16 s timeout.

If you can, please check 2.25 on your 14S: switch charge off/on from HA, and once with keep_alive: false.