HuBERT pre-training recipe

**Is your feature request related to a problem? Please describe:**
I have relied on fairseq for pre-training Wav2Vec2 and HuBERT models in the past; I see that fairseq2 includes [Wav2Vec2](https://github.com/facebookresearch/fairseq2/tree/main/recipes/wav2vec2) but seemingly lacks functionality for HuBERT at the moment.

**Describe the solution you would like:**
Include functionality for pre-training HuBERT in fairseq2, as it was available in [deprecated fairseq](https://github.com/facebookresearch/fairseq/tree/main/examples/hubert).

**Describe the alternatives you have considered:**
I want to be able to pre-train both Wav2Vec2 and HuBERT models within the same toolkit. Alternatively to fairseq2 I would consider switching to huggingface or PyTorch-native training, thus sadly missing out on fairseq2.

**Additional Context:**
HuBERT pre-training differs minimally from Wav2Vec2, only in pre-training objective and use of iterative refinement. It is still one of the most common architectures used for speech SSL and HuBERT models often outperform Wav2Vec2 on downstream tasks.