【FLCE】the bias in the previous layer

Nobody has claimed this yet.

I see the function LigerFusedLinearCrossEntropyLoss, I think bias and lin_weight are from the previous layer. If so, maybe `def forward(self, _input, target, lin_weight, bias=None):` is easy to understand.

```

def forward(self, lin_weight, _input, target, bias=None):
loss, z_loss = LigerFusedLinearCrossEntropyFunction.apply(
_input,
lin_weight,
target,
bias,
self.ce_weight,
self.ignore_index,
self.lse_square_scale,
self.label_smoothing,
self.reduction,
self.softcap,
self.return_z_loss,
self.accum_dtype,
)
if not self.return_z_loss:
return loss
return loss, z_loss
```

Start at LigerFusedLinearCrossEntropyLoss.forward and compare the proposed parameter order with the arguments passed to LigerFusedLinearCrossEntropyFunction.apply. Confirm whether lin_weight and bias represent the previous layer, then ensure the public signature and its call site use an unambiguous order.

Written by the indexing model from the issue text.