feat:Add MCP scenario module with tool poisoning, policy misconfig, detectors and CTF challenges

[nitinawari]

Summary

Adds an MCP scenario module so FinBot can run clean vs compromised tool environments, including return-value poisoning, policy misconfiguration, new detectors, and three CTF challenges.

Features added

• Tool output poisoning via output_append (return-value injection after a real tool call)
• Permission / policy misconfig (e.g. FinStripe max_payment, SystemUtils enabled tools)
• Benign vs compromised environment presets (Dark Lab apply for safe resets/policy; poison presets as preview → copy/paste; CLI can apply any preset)
• UnexpectedEndpointDetector for off-allowlist / blocked-host egress via network_request
• ToolOutputPoisoningDetector for output_append + PII exfil
• ToolDriftDetector wiring for MCP tool override / rediscovery drift
• Three new CTF challenges with OWASP / CWE / MITRE labels

Challenges added

Challenge ID	Title	Detector
data-exfil-output-poison	Return Value	ToolOutputPoisoningDetector
data-exfil-unexpected-egress	Unexpected Egress	UnexpectedEndpointDetector
recon-mcp-tool-drift	Tool Drift	ToolDriftDetector

Test plan

• pytest tests/unit/mcp/test_output_poisoning.py tests/unit/mcp/test_server_policy.py
• pytest tests/unit/scenarios/test_mcp_scenario_loader.py
• pytest tests/unit/ctf/test_mcp_supply_chain_detectors.py tests/unit/ctf/test_tool_drift_detector.py tests/unit/ctf/test_mcp_supply_chain_challenges.py

GSoC mapping

Week 9-11 (phase 4)

• Deliverable C : MCP-style scenario module (realism upgrade).