Design and implement a governance-rule sandboxing/resource-limiting execution model

[nurudeenheroic]

Summary

Implements a three-layer resource-limiting execution model for governance rule evaluation, preventing maliciously or accidentally crafted rule ASTs from degrading checkAccess() latency.

Layers of Defence

Layer	Mechanism	Enforced At	Purpose
1. Complexity limit	Weighted AST scoring (≤64)	Rule creation (validateRuleAST)	Rejects overly complex rules before storage
2. Depth limit	Recursion guard (≤10 levels)	Rule creation (validateRuleAST)	Prevents stack overflow
3. Wall-clock timeout	Date.now() deadline checks (5ms)	Evaluation (evaluateRuleWithBudget)	Bounds runtime of any accepted rule

Changes

packages/governance-engine/src/validator.ts

• Added computeComplexity() — weighted AST complexity scoring
• Added estimateComplexity() — pre-pass guard protecting recursive validator from pathologically wide×deep ASTs (e.g., 50¹⁰ worst case)
• Added RESOURCE_LIMITS export
• Added pre-pass complexity check (depth === 0) that rejects ASTs exceeding MAX_COMPLEXITY=64 before structural validation begins
• Updated validateRuleAST() to use the pre-pass guard

packages/governance-engine/src/evaluator.ts

• Added EvaluationOptions interface with timeoutMs
• Added evaluateRuleWithBudget() — budget-aware entry point with wall-clock deadline
• Added yield-point deadline check in evaluateNode() at each AST node entry
• Threaded deadline through all combinator functions (AND, OR, NOT, N_OF_M)
• evaluateRule() kept for backward compatibility

apps/access-api/src/policy/governanceRuleProvider.ts

• Switched from evaluateRule() to evaluateRuleWithBudget() with 5ms timeout
• Returns GOVERNANCE_TIMEOUT DENY code on timeout

tests

• 9 new unit tests for complexity scoring and budget-aware evaluation
• New load-test file with max-complexity AST (1000 evaluations bounded)

docs

• Added full resource-limiting execution model section to GOVERNANCE_ENGINE_IMPLEMENTATION.md

Acceptance Criteria

• validateRuleAST() rejects ASTs beyond complexity/depth threshold
• Evaluation is provably bounded in time (most complex valid AST completes in microseconds)
• Load test confirms max-complexity rule stays within budget (1000 iterations in ~55ms)
• Resource-limiting mechanism documented in GOVERNANCE_ENGINE_IMPLEMENTATION.md

Fixes #155