fix: Handle HTTP errors in FromResponseParts

[nerjs]

Fixed #5226
Fixed #5089

Problem

When decoding server function responses, if the return type goes through the FromResponseParts path (for example SetHeader<SetCookie> and similar types), the following implementation is used:

dioxus/packages/fullstack/src/magic.rs

Lines 235 to 251 in d01a820

	impl<T: FromResponse<R>, E, R> RequestDecodeResult<T, R> for &&&ServerFnDecoder<Result<T, E>> {
	type VerifyDecode = DecodeIsVerified;
	fn decode_client_response(
	&self,
	res: Result<R, RequestError>,
	) -> impl Future<Output = Result<Result<T, ServerFnError>, RequestError>> + Send {
	SendWrapper::new(async move {
	match res {
	Err(err) => Err(err),
	Ok(res) => Ok(T::from_response(res).await),
	}
	})
	}
	fn verify_can_deserialize(&self) -> Self::VerifyDecode {
	DecodeIsVerified
	}
	}

In the current implementation, this path ignores the HTTP status code (for example 4xx/5xx) and the response body containing error details. As a result, a non-2xx response can be incorrectly decoded as a success (for example Ok(SetHeader { data: None })) when the expected header is missing. This masks real server errors and makes the behavior inconsistent compared to the DeserializeOwned decoding path.