[chiconws]

Summary

Moves the SSH metric commands and their parsing from the Home Assistant integration into NanoKVMSSH, returning a typed model, and adds control of the non-Pro firmware's watchdog with a firmware check, so callers no longer gate it by version.

Changes

• get_metrics() returns an SSHMetrics dataclass: boot_time (UTC), cpu_temperature (°C), memory_total and storage_total (MiB, root filesystem), memory_used_percent, storage_used_percent and watchdog_enabled. The commands are the integration's (/proc/stat, thermal_zone0/temp, /proc/meminfo, df -k /).
• Unreadable values become None instead of failing the whole snapshot. Before, unexpected temperature output raised from a bare float(); a missing thermal zone now yields None too.
• get_watchdog_enabled() / set_watchdog_enabled(). kvm_system reboots the device when the video service misses its heartbeat (/tmp/nanokvm_wd) for more than 10 seconds, but only while /etc/kvm/watchdog exists. Support is detected by looking for that path in the kvm_system binary, which the Pro doesn't have. Without it, the getter returns None and the setter raises NanoKVMNotSupportedError.
• Docs. docs/ssh.md gains a metrics and watchdog section.

Notes

• The support check greps the 380 KB kvm_system binary on every watchdog call (about 0.3 s on the PCIe). It could be cached per connection if polling gets tight.
• Reconnecting stays with the caller, because NanoKVMSSH doesn't keep the password.

Behaviour changes

None. All methods are additions.

Testing

507 unit tests pass, 3 of them new. Against devices, every command run one by one, then through the new methods:

• NanoKVM PCIe 2.5.2: all metrics parsed. The watchdog read False; enabled, the heartbeat appeared within 1.3 s and it read True; disabled again after 3.9 s. Boot time unchanged afterwards, so no reboot.
• NanoKVM Pro 1.2.15: all metrics parsed. The watchdog read None, and set_watchdog_enabled(False) raised NanoKVMNotSupportedError without touching the file.