Provide teardown APIs in Fabric

Nobody has claimed this yet.

## 🚀 Feature

Right now, Fabric does not have any high-level APIs to free memory, unwrap models, etc.

### Motivation

Give the user device/strategy agnostic functions to call in their loops and training workflows when control is needed. Automate what can be automated.

### Pitch

- A way to unwrap layers that were wrapped with SyncBatchNorm (see also #13748)
- A way to free memory occupied by objects held within strategy

On the raw strategy/accelerator API level:
```py
accelerator.teadown_device()
model = strategy.teadown_module(model)
```

In the Fabric class
```py
fabric.free_memory()
model = fabric.teadown_module(model)
```

### Alternatives

- Have these methods only defined in the strategies.
- Keep the general purpose teardown() concept from PL which does many things, at the cost of flexibility in Lite.

### Additional context

Discussion started in https://github.com/Lightning-AI/lightning/pull/14662#discussion_r968623059

______________________________________________________________________

#### If you enjoy Lightning, check out our other projects! ⚡

- [**Metrics**](https://github.com/Lightning-AI/metrics): Machine learning metrics for distributed, scalable PyTorch applications.

- [**Lite**](https://pytorch-lightning.readthedocs.io/en/latest/starter/lightning_lite.html): enables pure PyTorch users to scale their existing code on any kind of device while retaining full control over their own loops and optimization logic.

- [**Flash**](https://github.com/Lightning-AI/lightning-flash): The fastest way to get a Lightning baseline! A collection of tasks for fast prototyping, baselining, fine-tuning, and solving problems with deep learning.

- [**Bolts**](https://github.com/Lightning-AI/lightning-bolts): Pretrained SOTA Deep Learning models, callbacks, and more for research and production with PyTorch Lightning and PyTorch.

- [**Lightning Transformers**](https://github.com/Lightning-AI/lightning-transformers): Flexible interface for high-performance research using SOTA Transformers leveraging PyTorch Lightning, Transformers, and Hydra.

cc @borda @tchaton @justusschock @awaelchli @carmocca

Start by reviewing the Fabric and raw strategy/accelerator APIs, then read the related SyncBatchNorm issue #13748 and the discussion linked from pull request #14662. Done would require agreed device-agnostic APIs for freeing strategy-held memory and unwrapping modules, with behavior validated across the relevant strategies.

Written by the indexing model from the issue text.