[Feature Request] Custom "chat" HF datasets

[chimezie]

The LoRa tuners's local datasets support the following data format:

{"messages": [{"role": "system", "content": "You are a helpful assistant."}, 
              {"role": "user", "content": "Hello."}, 
              {"role": "assistant", "content": "How can I assistant you today."}]}

Some HF datasets, such as the UltraFeedback dataset, used for Direct Preference Optimization (see: HF DPO trainer and #513) use a (json) data format such as the following:

[ { "content": "...", "role": "user" }, { "content": "...", "role": "assistant" } ]

To incorporate the use of such HF datasets, it would be helpful to to generalize the use of prompt_feature, text_feature, and completion_feature to include chat_feature, which indicates the HF dataset feature to use for the chat template structure.