`Trainer.validate()` after `Trainer.fit()` not working with FSDP and `auto_wrap_policy`

Nobody has claimed this yet.

### Bug description

I'm training large model with FSDP. Moreover, I'm using `_HYBRID_SHARD_ZERO2` sharding strategy and specify `configure_model` methon in Lightning Module.

At the moment I explicitly call validation after fit:
```python
trainer.fit(my_model, train_dataloader, val_dataloaders)
trainer.validate(my_model, val_dataloaders)
```

The fitting phase is fine, but when calling the validation I encounter this warning:
```python
A FSDP `auto_wrap_policy` is set, but the model is already wrapped. The policy will be ignored.
```
And after that the validation fails due to an error in `torch.Embedding` (RuntimeError: "weight" must be 2-D).

My guess is that after ignoring the policy, the model is sharded incorrectly and some nodes are running with the wrong embedding weights 🤷‍♂️ This is probably related to #18971, but I get the same error on the last version (2.1.2)

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

v2.1

### How to reproduce the bug

_No response_

### Error messages and logs

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

### Environment

Current environment

```
#- Lightning Component (e.g. Trainer, LightningModule, LightningApp, LightningWork, LightningFlow):
#- PyTorch Lightning Version (e.g., 1.5.0):
#- Lightning App Version (e.g., 0.5.2):
#- PyTorch Version (e.g., 2.0):
#- Python version (e.g., 3.9):
#- OS (e.g., Linux):
#- CUDA/cuDNN version:
#- GPU models and configuration:
#- How you installed Lightning(`conda`, `pip`, source):
#- Running environment of LightningApp (e.g. local, cloud):
```

### More info

_No response_

cc @awaelchli @carmocca

Start with the reported Trainer.fit() followed by Trainer.validate() sequence, using FSDP, _HYBRID_SHARD_ZERO2, auto_wrap_policy, and a LightningModule with configure_model. The issue provides no reproduction files or tests, so first isolate the failure and the model-wrapping state; done means validation completes without the auto_wrap_policy warning or the torch.Embedding shape error.

Written by the indexing model from the issue text.