Pre-release 1.4.4-pre1 🌞

A pre-release for testing. The headline is a rollback of the 1.4.3 limit-write change that turned out to cause a regression, plus a first pytest environment.

🩹 The 1.4.3 regression

• Revert "Send the full command on limit writes" (#1507, reverted in 6fc073d): routing output_limit / input_limit entity writes through the manager's power routines made every external write carry smartMode, acMode and a forced zero on the opposite limit. For anyone driving those entities from their own automations that alternated smartMode 0/1 on every command and parked the device in standby (#1521, reported by @mania-x, @Anzlinger, @rdeklerk04-source and @lexe). Limit writes are a plain single-property write again, as they were from 1.1.0 through 1.4.2.

  Consequence: the change that was meant to address #1505 (a device at 100% SoC ignoring valid limit writes) is out with it. That one was never confirmed fixed, and the recovery its reporter measured was a different sequence: set the limit to 0, wait for the device to start moving on its own, then set it back.

🩹 Fixes

• Manager mode select and ACE 1500 failing to load by @igosoft in #1527: guards an unusable property key, and a restored select value that is no longer in the option list now falls back instead of sticking.
• SF4000 Mix AC+ internal battery misidentified by @Rolf-Smit in #1529: packType 70 is the 8 kWh internal pack, which shares its serial prefix with the 0.96 kWh AB1000S.
• Do not report an unreachable zenSDK device as online (46fc9ad): a failed local poll returned an empty payload that was still applied, which pushed the device's last-seen forward and kept it online.
• Halve the zenSDK HTTP timeout to 2s (aa6e93b): the local read runs per device sequentially inside the control loop, so an unreachable device stalled it for the full timeout.

⚙️ Devices and configuration

• 4000 W and 5000 W fuse groups for the Mix series by @Rolf-Smit in #1556
• Refresh zenSDK devices on every cycle by @2tefan in #1542: previously the local read happened only once at startup.

🧪 Testing

• A pytest test environment by @krenk80 in #1540: first Python test coverage in CI, 14 scenario tests across device, number and manager, running on every push and pull request.

📖 Documentation

• Troubleshooting for the Mosquitto protocol error by @Kieft-C in #1524

🔧 Build and dependencies

• ruff 0.15.21 to 0.16.1 (#1573), colorlog 6.10.1 to 6.12.0 (#1553), bleak-retry-connector to >=4.6.3 (#1552), pip to >=26.2.1 (#1585), actions/stale 10 to 11 (#1572)

Full Changelog: 1.4.3...1.4.4-pre1