Harmonize Adapters and MultiWOZ w/ QLoRA

The LLama lobe adds QLoRA support via `peft` which isn't supported yet with our adapters. This model is used by the `MultiWOZ` recipe. Let's add QLoRA support and then harmonize the recipe with our adapters implementation: https://github.com/speechbrain/speechbrain/blob/develop/speechbrain/nnet/adapters.py