FSDP Strategy not working with bfloat16

Nobody has claimed this yet.

### Bug description

Flatten Param is breaking , "ValueError: Must flatten tensors with uniform dtype but got torch.bfloat16 and torch.float32". Models are loaded from huggingface in bf16 precision but fsdp couldn't autocast them to float32

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

v2.4

### How to reproduce the bug

_No response_

### Error messages and logs

ValueError: Must flatten tensors with uniform dtype but got torch.bfloat16 and torch.float32

### 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
@awaelchli @rohitgr7

_No response_

cc @lantiga

No source file, test, or reproducible example is named. Start by reproducing the FSDP Strategy failure with a Hugging Face model loaded in bfloat16, then trace the parameter-flattening and autocast path; done means the model runs without the mixed-dtype ValueError.

Written by the indexing model from the issue text.