Another MLP implementation along with multiplier support

### 🚀 The feature, motivation and pitch

Currently, LigerMLP modules only fuse swiglu/geglu computations together and leave matmuls untouched. These elementwise operations (including multiplier support #936 ) could be easily fused into matmul's epilogues. We can investigate the performance of this approach and see if we should adopt it.

TL;DR

Instead of
```python
gate_states = self.gate_proj(x)
up_states = self.up_proj(x)
intermidiate_states = LigerSiLUMulFunction.apply(gate_states , up_states)
return self.down_proj(intermidiate_states)
```
There are some other approaches worth exploring:

1. fuse activations (and multiplier) into gate_proj(x)
```python
up_states = self.up_proj(x)
intermidiate_states = LigerFusedLinearActMultiplierFunction.apply(x, self.gate_proj.weight, gate_multiplier, up_states)
return self.down_proj(intermidiate_states)
```
2. stack gate and up projections then put it into activation functions
```python
gate_up_states = self.gate_up_proj(x)
intermidiate_states = LigerSplitStatesActMultiplierFunction.apply(
gate_up_states,
config.hidden_act,
gate_multiplier,
up_states
)
return self.down_proj(intermidiate_states)
```
3. dual gemm with activations (and multiplier)
```python
intermidiate_states = LigerDualGemmActMulFuncion.apply(
x,
self.gate_proj.weight,
self.up_proj.weight,
config.hidden_act,
gate_multiplier,
)
return self.down_proj(intermidiate_states)
```

### Alternatives

_No response_

### Additional context

_No response_

Start by locating the LigerMLP modules and the existing fused SwiGLU/GeGLU implementation, then review multiplier support in issue #936. Compare the three proposed approaches with performance measurements; the work is done when an approach is selected, implemented, and its performance is documented.

Written by the indexing model from the issue text.