log traceback properly

[technillogue]

when an error happens, we log the traceback. this goes through the normal logging mechanism, which drops the log if there's more than one prediction. we can use _log(traceback.format_exc()) instead. also fix some minor formatting things.

[mattt]

@technillogue Can you share an example of what this looks like before and after the change?

[technillogue]

previously, two predictions at the same time, piped through jq '{error: .error, logs: .logs}

{
  "error": "",
  "logs": ""
}
{
  "error": "",
  "logs": "Traceback (most recent call last):\n stderrFile \"/home/sylv/r8/_cog/python/cog/server/worker.py\", line 240, in _handle_predict_error\n stderryield\n stderrFile \"/home/sylv/r8/_cog/python/cog/server/worker.py\", line 275, in _predict_async\n stderrasync for r in result:\n stderrFile \"/home/sylv/r8/_cog/python/test_predict.py\", line 69, in predict\n stderrraise Exception\n stderrException\n stderr"
}

now:

{
  "error": "",
  "logs": "Traceback (most recent call last):\n  File \"/home/sylv/r8/_cog/python/cog/server/worker.py\", line 240, in _handle_predict_error\n    yield\n  File \"/home/sylv/r8/_cog/python/cog/server/worker.py\", line 276, in _predict_async\n    async for r in result:\n  File \"/home/sylv/r8/_cog/python/test_predict.py\", line 69, in predict\n    raise Exception\nException\n"
}
{
  "error": "",
  "logs": "Traceback (most recent call last):\n  File \"/home/sylv/r8/_cog/python/cog/server/worker.py\", line 240, in _handle_predict_error\n    yield\n  File \"/home/sylv/r8/_cog/python/cog/server/worker.py\", line 276, in _predict_async\n    async for r in result:\n  File \"/home/sylv/r8/_cog/python/test_predict.py\", line 69, in predict\n    raise Exception\nException\n"
}