Add prediction field to ModelError

[mattt]

This PR extends #325 to add the prediction object itself to ModelError, as opposed to just its ID. This makes it convenient to introspect logs and other information to determine how to handle the failure.

import replicate
from replicate.exceptions import ModelError

try:
  output = replicate.run("stability-ai/stable-diffusion-3", { "prompt": "..." })
except ModelError as e
  if "(some known issue)" in e.prediction.logs:
    pass

  print("Failed prediction: " + e.prediction.id)