fix(panasonic): pass auto_power_on through to the change request builder

[Tonttu84]

2026.8.7 added the auto_power_on option (#495), but setting it to false has no effect — a settings write to a powered-off unit still turns it on.

panasonic/__init__.py passes the option to the client correctly. PanasonicDeviceCoordinator.get_change_request_builder then constructs its own builder without it:

def get_change_request_builder(self) -> ChangeRequestBuilder:
    return ChangeRequestBuilder(self.device)

ChangeRequestBuilder.__init__ defaults to auto_power_on=True, so _ensure_powered_on() adds operate: On to the request. That also defeats the other half of the design: set_device_raw() buffers a change for an off device unless the request is turning it on — and this one always is.

ApiClient already exposes the right factory: new_change_request(device), documented as "Create a ChangeRequestBuilder wired to this client's auto_power_on setting".

-return ChangeRequestBuilder(self.device)
+return self._api_client.new_change_request(self.device)

Tested

On a real CS-HZ25XKE, integration 2026.8.7, with auto_power_on: false set on the entry — not in the devcontainer:

step	before	after
unit commanded off	off	off
setpoint written while off	heat — woken	off — change buffered
explicit hvac_mode: heat	heat	heat, and the buffered setpoint applied

The third row is the one worth reviewing: set_hvac_mode uses _force_powered_on(), so an explicit mode command still powers the unit on as intended. Only the implicit wake-on-settings-change is suppressed, which is what the option is for.

Notes

• No requirements change: new_change_request is in aio-panasonic-comfort-cloud==2026.8.9, already pinned by the manifest.
• The ChangeRequestBuilder import stays in use as a type annotation.
• coordinator.py does not currently pass black --check, but that predates this change and black leaves this line alone, so the file is left otherwise untouched rather than reformatted in a one-line fix.

This may also be relevant to #378 and #474, which describe a unit coming back on after being turned off — the comment on #378 identifies a setpoint write as the trigger. I haven't reproduced either reporter's setup, and neither is on a version carrying this option, so I'm not claiming to close them.