Fix off-by-one in is_partial_stop causing false positives

[Chessing234]

Summary

is_partial_stop checks whether the tail of the streaming output is a partial match for a stop string. The loop range range(0, min(len(output), len(stop_str))) has two bugs:

1. i=0 on first iteration: output[-0:] in Python evaluates to output[0:] — the entire string, not an empty suffix. So stop_str.startswith(output) is tested, returning True whenever the full output happens to be a prefix of the stop string. This incorrectly suppresses streaming chunks.

2. Exclusive upper bound: The largest valid suffix (of length min(len(output), len(stop_str))) is never checked.

Fix: Change range(0, ...) to range(1, ... + 1) so the loop checks suffixes of length 1 through min(len(output), len(stop_str)).

You can verify in a Python REPL:

>>> s = "hello"
>>> s[-0:]
'hello'  # entire string, not empty
>>> s[-1:]
'o'      # correct: last 1 char

Test plan

• Verified output[-0:] returns full string in Python REPL
• Existing tests in tests/test_utils.py should continue to pass

🤖 Generated with Claude Code