fix: extract multi-digit state indices in extract_state_value_from_output

[Osamaali313]

Problem

extract_state_value_from_output is used by Role._think to read the next state index from an LLM's reply (role.py):

pattern = r"(?<!-)[0-9]"
matches = re.findall(pattern, content, re.DOTALL)
matches = list(set(matches))
state = matches[0] if len(matches) > 0 else "-1"

Two issues:

1. The pattern matches a single digit, so a state index >= 10 is split into separate digits. For content == "10" the result is "1" or "0" — never "10".
2. list(set(matches)) makes the choice order-dependent on set hashing, which varies with PYTHONHASHSEED across processes, so the returned digit is nondeterministic.

Because the caller only validates int(next_state) in range(-1, len(self.states)), a wrong-but-in-range digit passes silently, so any role with 10+ states can transition to the wrong state (or loop) without error.

Fix

Match the first optionally-negative, multi-digit integer and drop the set() dedup:

pattern = r"-?\d+"
matches = re.findall(pattern, content, re.DOTALL)
state = matches[0] if len(matches) > 0 else "-1"

Single-digit outputs and the "-1" no-op state are unchanged; multi-digit indices are now returned whole and deterministically.

Test

Adds test_extract_state_value_from_output covering "0", " 3\n", "-1", "10", and "The next state is 12.". Without the fix the "10" and "12" cases fail (return a single digit).

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.