fix: Remove eval() from workflow condition evaluation

[Joshua-Medvinsky]

Summary

• Removes unsafe eval() call used to evaluate workflow condition strings
• Forces callers to supply a proper callable (e.g. lambda) instead of a raw expression string
• Eliminates remote code execution risk when condition values originate from user-supplied input

Vulnerability

python_a2a/workflow.py evaluated the condition parameter with Python's eval() if it was not callable:

self.condition_func = lambda: eval(self.condition, {"agent": self.agent})

An attacker who controls the condition argument (e.g. via a serialised workflow definition or API call) can pass any Python expression — including __import__('os').system('...') — resulting in arbitrary code execution with the privileges of the server process.

Severity: Critical
CWE: CWE-94 (Code Injection)

Fix

if not callable(self.condition):
    raise ValueError("Condition must be a callable, not a string. Use lambda instead.")

Callers that previously passed strings must migrate to lambdas:

# Before (unsafe)
Workflow(condition="some_expression")

# After (safe)
Workflow(condition=lambda: some_expression)

Test plan

• Confirm ValueError is raised when a string condition is passed
• Confirm existing callable conditions continue to work
• No regression in existing workflow tests

[Joshua-Medvinsky]

Closing — duplicate of #89 which was submitted moments later with the same fix. Sorry for the noise.