KeyError: 'Trying to restore optimizer state but checkpoint contains only the model. This is probably due to `ModelCheckpoint.save_weights_only` being set to `True`.' But optim_cfg is in model

Nobody has claimed this yet.

### Bug description

I want to train from several epochs in and I can't implement correctly. I tried training by doing the following:

```
model = DeepSpeech.load_from_checkpoint(loaded_checkpoint_path)
trainer = pl.Trainer(max_epochs=max_epoch,
callbacks =[checkpoint_callback_step,checkpoint_callback_epoch],
precision = 32,accelerator = 'auto',
gradient_clip_val= 400,
accumulate_grad_batches= 1,
enable_progress_bar=False,
log_every_n_steps = 250,
logger=neptune_logger)

data_module = DeepSpeechDataModule(data_cfg,augment_cfg)
if train_from_checkpoint:
trainer.fit(model, datamodule=data_module)
```

And it treats this as epoch 0 but has the weights loaded.

If instead I do:

```
model = DeepSpeech(
labels=data_cfg.vocab, #labels,
model_cfg=model_cfg,
optim_cfg=optim_cfg,
precision=32, #DeepSpeechTrainerConf.precision,
spect_cfg=spect_cfg
)
trainer.fit(model, datamodule=data_module,ckpt_path=loaded_checkpoint_path)
```

I get the error:
KeyError: 'Trying to restore optimizer state but checkpoint contains only the model. This is probably due to `ModelCheckpoint.save_weights_only` being set to `True`.'

How do I get the saved weights and optimizer settings?

The ModelCheckpoint is:
```
checkpoint_callback_epoch = ModelCheckpoint(
dirpath=check_path,# Directory to save checkpoints
filename=f'unidirDeepSpeech_{save_targ_string}_{{epoch:02d}}-{{val_wer:.2f}}_c{{val_cer:.2f}}_{date_string}',
save_top_k=3, # Save the top 3 models
monitor='val_cer', # Metric to monitor for checkpointing
mode='min', # Minimize the monitored metric (e.g., lower val_loss is better)
save_weights_only = 'False'
)
```

Thanks for any help.

### What version are you seeing the problem on?

v2.2

### How to reproduce the bug

```python
pl.__version__ = '2.2.0.post0'
```

### Error messages and logs

```
# Error messages and logs here please
```

### Environment

Current environment

```
#- PyTorch Lightning Version (e.g., 2.4.0):
#- PyTorch Version (e.g., 2.4):
#- Python version (e.g., 3.12):
#- OS (e.g., Linux):
#- CUDA/cuDNN version:
#- GPU models and configuration:
#- How you installed Lightning(`conda`, `pip`, source):
```

### More info

_No response_

cc @lantiga

Start with the ModelCheckpoint configuration and the trainer.fit calls shown in the report, comparing load_from_checkpoint with ckpt_path restoration. Reproduce the KeyError on version 2.2.0.post0 and determine whether the checkpoint contains optimizer state; done means the report has a minimal reproduction and a verified restoration path.

Written by the indexing model from the issue text.