[refactor] feat: separate data, batch and metric with clear variable names

[tongyx361]

What does this PR do?

This PR separates data, batch and metric with clear variable names to avoid reusing the same name data for different variables.

ChangeLog:

• Rename variables according to loop level and class
  • loop level: (no prefix) -> mini -> micro
  • class:
    • DataProto: data
    • TensorDict: batch
    • dict for metrics: metric_data
• Simplify code for multi-modal data loading:

non_tensor_select_keys = ["multi_modal_inputs"] if "multi_modal_inputs" in data.non_tensor_batch.keys() else []

Usage:

No change.

Before submitting

• Did you read the Contribute Guide and finish the code format check?
• Did you make sure to update the documentations with your changes in the docs especially for breaking config etc?
• Did you write any test cases if neccessary? Please add CI tests to your new feature.

Additional Info:

• Issue Number: none
• Training: FSDP
• Inference: none