Add max_tokens parameter to all tools

[iiivanova]

Summary

All tools currently hardcode max_tokens=1024. For longer analyses the output can get cut off. Add an optional max_tokens parameter to each tool that defaults to 1024 to preserve existing behavior.

Affected functions

• evaluate()
• plan()
• redteam()
• ask()

Expected usage

# default behavior unchanged
print(evaluate(text))

# override for longer output
print(evaluate(text, max_tokens=2048))

Implementation notes

• Add max_tokens: int = 1024 to each tool's function signature
• Pass through to create_message() in providers.py
• Update tests to cover non-default values