Add BMUF Multi-GPU Training Method

Nobody has claimed this yet.

## 🚀 Feature

### Motivation

Implements incremental block distributed data parallelism similar to https: ieeexplore.ieee.org/document/7472805.
This can mitigate the performance loss caused by multi-card training.

### Pitch
This can be found in the [Fairseq implementation](https://github.com/facebookresearch/fairseq/blob/b5a039c292facba9c73f59ff34621ec131d82341/fairseq/optim/bmuf.py).

### Additional context

______________________________________________________________________

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

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

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

cc @borda @awaelchli @rohitgr7 @akihironitta

Start by reviewing the linked Fairseq implementation at fairseq/optim/bmuf.py and the surrounding PyTorch Lightning training and distributed-training entry points. Determine how BMUF should integrate with the existing multi-GPU workflow and what tests cover distributed optimization. Done means BMUF training is supported with documented behavior and passing coverage for the new method.

Written by the indexing model from the issue text.