[Codex] fix: harden AST sandboxing in logic executor to prevent sandbox escapes

[Micsi]

Source PR

• Fork PR: fix: harden AST sandboxing in logic executor to prevent sandbox escapes Micsi/openbridgeserver#65

Motivation

• The logic engine allowed user-supplied formula and script strings to reach eval/exec which enabled attribute-traversal sandbox escapes and authenticated RCE.
• The goal is to minimally harden the executor so formulas and scripts cannot use dangerous AST constructs or object-graph traversal while preserving intended math/script features.

Description

• Added GraphExecutor._validate_formula_ast which allows only a constrained set of expression AST node types and blocks attribute access except safe math.<name> lookups.
• Added GraphExecutor._validate_script_ast which blocks dangerous script constructs (imports, defs, class/async/lambda/with/try/global/nonlocal/raise/delete/yield/await) and disallows attribute traversal except safe math.<name> access.
• Integrated these validators into _safe_eval (validate parsed formula AST before eval) and _run_script (validate parsed script AST and compile before exec).
• Updated unit test tests/unit/test_executor.py to assert attribute traversal is blocked instead of documenting the previous unsafe behavior.

Testing

• Ran pytest -q tests/unit/test_executor.py and the test file passed with all tests green (192 passed, 1 warning).
• The change is minimal, focused on AST validation in obs/logic/executor.py and the associated unit test update in tests/unit/test_executor.py.

---

Codex Task

[Micsi]

Zusatzupdate aus dem Security-Review (TDD-basiert):

• Formula-Sandbox so nachgezogen, dass math.<name> in Formeln tatsächlich erreichbar ist (z. B. math.sqrt(x)), ohne die Sperre für Objekt-Graph-Traversal zu lockern.
• Dunder-Zugriffe bleiben blockiert (math.__dict__ in Formula und Script).
• Zusätzlicher Negativtest für disallowte Formula-Syntax (lambda) ergänzt.
• Tests erweitert und grün: pytest -q tests/unit/test_executor.py -> 196 passed.
• Lint grün auf den geänderten Dateien: ruff check obs/logic/executor.py tests/unit/test_executor.py.
• Coverage im betroffenen Modul verbessert auf 70% (obs/logic/executor.py), Diff-Coverage für geänderte Executor-Zeilen: 32/32.