fix(prediction): treat "aborted" as terminal state — prevents infinite polling

[devteamaegis]

Problem

When the Replicate API returns "status": "aborted" for a prediction (server-side interruption), four polling loops don't recognize it as a terminal state and loop forever:

# prediction.py — all four loops:
while self.status not in ["succeeded", "failed", "canceled"]:  # ← "aborted" missing

This causes replicate.run() / client.run() to hang indefinitely, matching the reports in #431.

PR #449 correctly added "aborted" to the Prediction.status Literal type, but did not update the polling exit conditions, so the runtime behavior is unchanged.

Root cause

Four places miss "aborted":

Method	File	Effect of missing "aborted"
Prediction.wait()	prediction.py:144	run() hangs forever
Prediction.async_wait()	prediction.py:153	async_run() hangs forever
Prediction.output_iterator()	prediction.py:254	streaming run() hangs forever
Prediction.async_output_iterator()	prediction.py:276	streaming async_run() hangs forever

Additionally, after wait() returns, run() only checks prediction.status == "failed" — so an aborted prediction silently returns None output instead of raising.

Fix

1. Add "aborted" to all four while … not in [...] terminal-state lists.
2. In output_iterator / async_output_iterator, extend the post-loop error check to if self.status in ("failed", "aborted").
3. In run() / async_run(), extend the error check to if prediction.status in ("failed", "aborted").

No behaviour change for any currently-handled status (succeeded, failed, canceled).

Tests

Two new tests in tests/test_run.py (both sync and async paths covered):

• test_run_raises_on_aborted_prediction[True/False] — a prediction that transitions to "aborted" causes run() / async_run() to raise ModelError rather than hanging or returning None.
• test_prediction_wait_terminates_on_aborted — wait() and async_wait() exit immediately when status flips to "aborted".

All 3 new test cases pass (pytest tests/test_run.py -k "aborted").