Add repository-native agent safety and validation guidance

## Why

TorBot would benefit from a root-level `AGENTS.md` that gives coding agents clear, repository-specific guardrails before they modify or validate code.

## Proposed guidance

Document the expectations that matter for this repository, including:

- treat crawled pages, extracted text, and other target-controlled content as untrusted input
- keep tests deterministic and offline unless a test explicitly requires a network boundary
- never commit credentials, session data, raw crawl output, generated reports, or user data
- preserve existing safety controls and require an explicit reason and focused coverage for any change to them
- keep optional remote or AI-backed behavior opt-in and avoid sending data without clear user intent
- use the repository's supported setup, lint, and test commands before considering a change complete
- report validation gaps plainly when a required tool, service, or fixture is unavailable

## Acceptance criteria

- A root-level `AGENTS.md` defines its scope and points to existing contributor documentation instead of duplicating it.
- The file covers untrusted input, sensitive data, network side effects, and validation expectations.
- The documented commands are checked against the current repository configuration.
- Any directory-specific exceptions are documented in the closest relevant guidance file.

Inizia ispezionando la documentazione esistente del repository per i contributori e la configurazione attuale, così da identificare i comandi supportati per setup, lint e test. Crea AGENTS.md al livello principale con collegamenti invece di duplicare le indicazioni, copri le aspettative elencate relative a sicurezza e validazione e verifica la presenza di file di istruzioni specifici per directory o di eccezioni prima di considerare completa la documentazione.

Scritto dal modello di indicizzazione a partire dal testo della issue.