Rewrite new_zeros, new_ones, new_full decomp with aten.full (#84946)
We should **NOT** introducing non-functional op for decomps of functional op.
For example
```
make_fx(functionalize(lambda x: x.new_zeros(3)), decomposition_table=decomposition_table)(x)
```
is producing
```
def forward(self, x_1):
empty = torch.ops.aten.empty.memory_format([3, 4], dtype = torch.float32, layout = torch.strided, device = device(type='cpu'), pin_memory = False)
zero_ = torch.ops.aten.zero_.default(empty); empty = None
return zero_
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/84946
Approved by: https://github.com/ngimel
