Add streaming support for tool responses

[iiivanova]

Feature request

Add streaming support so output is returned as it's generated rather than waiting for the full response.

Why

Currently all tool calls block until the complete response is ready, which can take 10-20 seconds for longer analyses. Streaming would show output word by word as it arrives, making the experience feel much more responsive - especially for evaluate() and redteam() which produce long outputs.

Example usage

for chunk in evaluate("my analysis", stream=True):
    print(chunk, end="", flush=True)

What's needed

• Add an optional stream=True parameter to each tool
• Update the provider clients in providers.py to support streaming for Anthropic, OpenAI, and Gemini
• Default behavior (no streaming) remains unchanged