Documentation improvements and make toolkit easy to run

[Shikhar-S]

1. improve documentation

• contributing guidelines mention some pre-hoook from the old template
• switch comments and default configs from vaani to a dataset that can be downloaded from hf
• mentions using callbacks for metrics, this is only during training mode. make this clear.
• mention clearly that intrinsic datasets are not yet uploaded to hf

2. pfer computation should directly load jsonl without the intermittent jsonl2json script.
3. upload intrinsic datasets to hf in kaldi format. reuse kaldi dataset for intrinsic evals

Thanks @SanderGi for bringing up these issues.