fix: outputs order is wrong when more than 1 outputs

[mhchia]

What's wrong?

With computation returning more than 1 outputs, we get the wrong order of outputs. For example, the computation like this

def computation(state: State, args: list[torch.Tensor]):
    x = args[0]
    y = args[1]
    return state.mean(x), state.median(y)

results in outputs like

51.5, 1, 46.25

where 1 is the boolean for verifiers to check the result is accurate enough. It is wrong since 1 should always be the first return value of a computation

The issue is that we return is_precise_aggregated, op.result+(x[0]-x[0])[0][0] for the last operation. However, the return value of a computation is not necessarily the return value of the last operation. For example,

def computation(state: State, args: list[torch.Tensor]):
    x = args[0]
    y = args[1]
    return state.mean(x), state.median(y)

the last operation is state.median(y) but the return value is state.mean(x), state.median(y). In our original approach, we'll return res_mean, is_precise_aggregated, res_median.

How it's fixed?

Instead of returning the is_precise_aggregated in State.call_ops, we do it in the Model.forward

def forward(self, *x: list[torch.Tensor]) -> tuple[IsResultPrecise, torch.Tensor]:
            result = computation(state, x)
            is_computation_result_accurate = state.bools[0]()
            for op_precise_check in state.bools[1:]:
                is_op_result_accurate = op_precise_check()
                is_computation_result_accurate = torch.logical_and(is_computation_result_accurate, is_op_result_accurate)
            return is_computation_result_accurate, result