lr_find crashes with Deepspeed

Nobody has claimed this yet.

## 🐛 Bug

When running Trainer.tuner.lr_find using strategy="deepspeed_stage_2", the procedure fails with following error:
```
AttributeError: 'BoringModel' object has no attribute 'save_checkpoint'
```

Full traceback

```
Traceback (most recent call last):
File "src/tmp.py", line 161, in
run()
File "src/tmp.py", line 154, in run
lr_finder = trainer.tuner.lr_find(
File "/root/envs/star/lib/python3.8/site-packages/pytorch_lightning/tuner/tuning.py", line 185, in lr_find
result = self.trainer.tune(
File "/root/envs/star/lib/python3.8/site-packages/pytorch_lightning/trainer/trainer.py", line 1099, in tune
result = self.tuner._tune(model, scale_batch_size_kwargs=scale_batch_size_kwargs, lr_find_kwargs=lr_find_kwargs)
File "/root/envs/star/lib/python3.8/site-packages/pytorch_lightning/tuner/tuning.py", line 53, in _tune
result["lr_find"] = lr_find(self.trainer, model, **lr_find_kwargs)
File "/root/envs/star/lib/python3.8/site-packages/pytorch_lightning/tuner/lr_finder.py", line 232, in lr_find
trainer.save_checkpoint(str(save_path))
File "/root/envs/star/lib/python3.8/site-packages/pytorch_lightning/trainer/trainer.py", line 1913, in save_checkpoint
self.checkpoint_connector.save_checkpoint(filepath, weights_only)
File "/root/envs/star/lib/python3.8/site-packages/pytorch_lightning/trainer/connectors/checkpoint_connector.py", line 478, in save_checkpoint
self.trainer.training_type_plugin.save_checkpoint(_checkpoint, filepath)
File "/root/envs/star/lib/python3.8/site-packages/pytorch_lightning/plugins/training_type/deepspeed.py", line 751, in save_checkpoint
self.deepspeed_engine.save_checkpoint(filepath, client_state=checkpoint)
File "/root/envs/star/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1177, in __getattr__
raise AttributeError("'{}' object has no attribute '{}'".format(
AttributeError: 'BoringModel' object has no attribute 'save_checkpoint'
```

### To Reproduce

```
import os

import torch
from torch.utils.data import DataLoader, Dataset

from pytorch_lightning import LightningModule, Trainer

class RandomDataset(Dataset):
def __init__(self, size, length):
self.len = length
self.data = torch.randn(length, size)

def __getitem__(self, index):
return self.data[index]

def __len__(self):
return self.len

class BoringModel(LightningModule):
def __init__(self):
super().__init__()
self.lr = 0.1
self.layer = torch.nn.Linear(32, 2)

def forward(self, x):
return self.layer(x)

def training_step(self, batch, batch_idx):
loss = self(batch).sum().abs()
self.log("train_loss", loss)
return {"loss": loss}

def validation_step(self, batch, batch_idx):
loss = self(batch).sum().abs()
self.log("valid_loss", loss)

def test_step(self, batch, batch_idx):
loss = self(batch).sum().abs()
self.log("test_loss", loss)

def configure_optimizers(self):
return torch.optim.SGD(self.layer.parameters(), lr=self.lr)

def run():
train_data = DataLoader(RandomDataset(32, 4), batch_size=8)
val_data = DataLoader(RandomDataset(32, 4), batch_size=8)
test_data = DataLoader(RandomDataset(32, 4), batch_size=8)

model = BoringModel()
trainer = Trainer(
accelerator="gpu", devices=1,
strategy="deepspeed_stage_2",
precision=16,
default_root_dir=os.getcwd(),
limit_train_batches=1,
limit_val_batches=1,
limit_test_batches=1,
num_sanity_val_steps=0,
max_epochs=1,
enable_model_summary=False,
)

lr_finder = trainer.tuner.lr_find(
model, train_dataloaders=train_data, val_dataloaders=val_data)

print(lr_finder.suggestion())

if __name__ == "__main__":
run()
```

### Expected behavior

Learning rate finder should work with deepspeed or else raise an exception stating that deepspeed is not supported.

### Environment

* CUDA:
- GPU:
- Tesla V100-SXM2-32GB
- available: True
- version: 11.3
* Packages:
- numpy: 1.21.0
- pyTorch_debug: False
- pyTorch_version: 1.10.1+cu113
- pytorch-lightning: 1.5.10
- tqdm: 4.62.3
* System:
- OS: Linux
- architecture:
- 64bit
- processor: x86_64
- python: 3.8.10
- version: #1 SMP Tue Apr 13 18:55:45 EDT 2021

### Additional context

It is valuable to be able to run the lr_finder with deepspeed for configurations where large batch sizes matter (e.g., contrative learning).

cc @akihironitta @borda @rohitgr7 @SeanNaren @awaelchli

Start with pytorch_lightning/tuner/lr_finder.py and the deepspeed save_checkpoint path in plugins/training_type/deepspeed.py. Run the supplied reproduction with strategy="deepspeed_stage_2" and trace how lr_find saves state. Done means the learning-rate finder works with DeepSpeed or raises an explicit unsupported-feature exception.

Written by the indexing model from the issue text.