Regression in download, `openai api fine_tunes.results` behavior

[thejaminator]

To reproduce, on the older openai==0.25.0

openai api fine_tunes.results -i <job_id_here>

You'll get the output of the actual CSV file

E.g.

step,elapsed_tokens,elapsed_examples,training_loss,training_sequence_accuracy,training_token_accuracy
1,22048,32,0.011019098009307718,0.0,0.0
1,32288,32,0.007613447834811513,0.0,0.0

On the latest openai==0.26.2

You get instead

{
  "object": "file",
  "id": "file-66NnpqAXzw3In27fROPIU03P",
  "purpose": "fine-tune-results",
  "filename": "compiled_results.csv",
  "bytes": 386756,
  "created_at": 1674504271,
  "status": "processed",
  "status_details": null
}

This is probably due to a change in behavior in download of /api_resources/file.py. It probably affects everyone who depends on that download method, so i'm guessing wandb users who are syncing their runs too?

See https://github.com/openai/openai-python/pull/146/files#r1084884786

Probably related to #185, #183