Model Parallelism and accelerate's usage of DDP aren't compatible

### System Info

```Shell
- `Accelerate` version: 0.18.0
- Platform: Linux-5.4.0-124-generic-x86_64-with-glibc2.31
- Python version: 3.9.12
- PyTorch version (GPU?): 1.12.0 (True)
- `Accelerate` default config:
- compute_environment: LOCAL_MACHINE
- distributed_type: MULTI_GPU
- mixed_precision: no
- use_cpu: False
- num_processes: 16
- machine_rank: 0
- num_machines: 16
- main_process_ip: 192.168.1.1
- main_process_port: 8080
- rdzv_backend: static
- same_network: False
- main_training_function: main
- downcast_bf16: no
- tpu_use_cluster: False
- tpu_use_sudo: False
- tpu_env: []
```

### Information

- [ ] The official example scripts
- [X] My own modified scripts

### Tasks

- [ ] One of the scripts in the examples/ folder of Accelerate or an officially supported `no_trainer` script in the `examples` folder of the `transformers` repo (such as `run_no_trainer_glue.py`)
- [X] My own task or dataset (give details below)

### Reproduction

If I use model parallelism (for example using huggingface parallelize), and I'm using accelerate with a standard multi-GPU environment (that uses DDP), then when I prepare the model I get the following error:

```
File "/private/home/raileanu/new-rlvsil/rlvsil/experiment_accel.py", line 689, in main
model, optimizer, lr_scheduler, *prepared_dataloaders = accelerator.prepare(
File "/private/home/raileanu/.conda/envs/rob/lib/python3.9/site-packages/accelerate/accelerator.py", line 1122, in prepare
result = tuple(
File "/private/home/raileanu/.conda/envs/rob/lib/python3.9/site-packages/accelerate/accelerator.py", line 1123, in
self._prepare_one(obj, first_pass=True, device_placement=d) for obj, d in zip(args, device_placement)
File "/private/home/raileanu/.conda/envs/rob/lib/python3.9/site-packages/accelerate/accelerator.py", line 977, in _prepare_one
return self.prepare_model(obj, device_placement=device_placement)
File "/private/home/raileanu/.conda/envs/rob/lib/python3.9/site-packages/accelerate/accelerator.py", line 1202, in prepare_model
model = torch.nn.parallel.DistributedDataParallel(
File "/private/home/raileanu/.conda/envs/rob/lib/python3.9/site-packages/torch/nn/parallel/distributed.py", line 571, in __init__
self._log_and_throw(
File "/private/home/raileanu/.conda/envs/rob/lib/python3.9/site-packages/torch/nn/parallel/distributed.py", line 674, in _log_and_throw
raise err_type(err_msg)
ValueError: DistributedDataParallel device_ids and output_device arguments only work with single-device/multiple-device GPU modules or CPU modules, but got device_ids [0], output_device 0, and module parameters {device(type='cuda', index=0), device(type='cuda', index=1)}.
```

I think this is because in line https://github.com/huggingface/accelerate/blob/2708c1ae31f5c32a0715780c2244c8f24ba1cfc3/src/accelerate/accelerator.py#L1222 it initialises the DDP model by setting device_ids and output_device, whereas these should both be set to `None` if using model parallelism.

You should be able to reproduce this on a 4-gpu machine with something like the following:

```python
from transformers import GPTJForCausalLM
import accelerate

model = GPTJForCausalLM.from_pretrained("EleutherAI/gpt-j-6B")
device_map = {
0: [0, 1, 2, 3, 4, 5, 6],
1: [7, 8, 9, 10, 11, 12, 13],
2: [14, 15, 16, 17, 18, 19, 20],
3: [21, 22, 23, 24, 25, 26, 27],
}
model.parallelize(device_map)

accelerator = accelerate.Accelerator()

model = accelerator.prepare(model)
```

I'm currently getting around this by wrapping the model in DDP myself with the correct arguments, and then doing `accelerator._models.append(model)`.

### Expected behavior

I'd expect accelerate's usage of DDP to be compatible with naïve model parallelism, as DDP is compatible with it.

I think the fix would be to adjust https://github.com/huggingface/accelerate/blob/2708c1ae31f5c32a0715780c2244c8f24ba1cfc3/src/accelerate/accelerator.py#L1222 such that if the model has parameters on multiple devices, or the hf_device_map uses multiple devices, (or maybe the user passes an explicit parameters saying they're using model parallelism), the DDP initialisation doesn't set `device_ids` and `output_device`. I'd be happy to submit a PR to make that change if that seems reasonable.