Security hardening, refactor and robustness fixes across AI providers, tools and CLI

[prithvee07]

Motivation

• Improve security and privacy at the boundary between external tools and AI/persistence by redacting secrets, enforcing file permissions and avoiding accidental leakage of sensitive command arguments.
• Harden parsing and validation to prevent common injection and resource-exhaustion attacks (e.g. large or malicious XML, unsafe flags), and make provider/tool integrations more robust.
• Clean up imports, fix minor CLI/formatting issues and tighten dependency bounds for more predictable builds.

Description

• Add central redaction utilities via utils.redaction.redact_sensitive_text and apply redaction to tool outputs and session persistence, and ensure BaseTool never exposes raw executable command with secrets by logging a sanitized safe_command instead.
• Secure session and report files by using restrictive file permissions (0o600 for files, 0o700 for directories) when saving state and reports, and use safe open flags to avoid symlink following.
• Harden parsers and validators: tools/nmap.py now rejects DTD/entity-containing or oversized XML inputs; ScopeValidator tightens blacklist logic and DNS resolution checks to prevent DNS-rebinding and blocks non-global addresses; many tool parsers and wrappers were made more resilient against malformed output.
• Refactor and tidy code style across the project: rearranged imports, consistent string quoting, trailing commas, re-ordered/cleaned ai.prompt_templates.__init__, provider implementations updated to import LangChain message classes consistently, and dependency bounds added to pyproject.toml.

Testing

• Ran the unit test suite with pytest (updated/added tests include tests/test_redaction.py, enhancements to tests/test_memory.py, tests/test_nmap_tool.py, tests/test_base_tool.py, and others) and verified the test run completed successfully.
• New tests exercise redaction, secure file permissions, nmap XML protections, and command sanitization and passed.
• No failing automated tests remained after the changes.

---

Codex Task