Add Testing Protocol to AGENTS.md

Description

Following up on the "Anti-Corruption Protocol" (Section 8), we need to establish clear rules for how agents test code to avoid cluttering the environment or relying on dangerous shell scripts.

LLMs often default to writing throwaway test.js files or using cat << EOF > tmp.js to verify behavior. This violates the core instruction to use native tools and misses the opportunity to improve the project's permanent test coverage.

Proposed Solution

Add a new Section 9 to AGENTS.md titled "Testing and Validation Protocol".

This section should mandate:

1. Micro-Benchmarking (V8 physics): Agents are allowed to use run_shell_command with node -e '...' for fast, ephemeral performance validation (e.g., scoping let vs var).
2. Framework Logic: Agents are FORBIDDEN from using throwaway bash scripts (cat > test.js) or temporary files to test Neo.mjs logic.
3. Permanent Coverage: All framework behavioral validation must be written directly into the appropriate Playwright unit test file (e.g., test/playwright/unit/data/Store.spec.mjs) using write_file or replace.