docs: add tutorial — Multi-Technique Guardrail Bypass Evaluation

[onurcangnc]

Adds a new field-observation tutorial documenting four guardrail-bypass technique families (authority/context reframing, obfuscation-based filter evasion, DAN-class persona injection, business-framing reframe) evaluated across a five-stage black-box sequence against a single chatbot deployment. Includes OWASP LLM01:2025 mapping, MITRE ATLAS references, and NIST CSF 2.0 alignment via the OWASP LLM Top 10 v2.0 OLIR crosswalk. Methodology explicitly flags a confounded comparison in Stage 2 and marks all mechanism claims as black-box observations rather than confirmed architecture. Personal names sanitized to {TARGET_NAME}; harm-enabling outputs and functional exploit code intentionally not reproduced.