docs: add related-work and deployment-contexts

Two new docs that the kit was missing.

related-work.md positions the kit explicitly against the technical
AI-safety landscape that has grown around the same problem:

- System prompts and behavioral defaults
- Constitutional AI (Bai et al. 2022)
- RAG with citation enforcement
- RLHF / DPO tuning against sycophancy

And against the adjacent classes that get conflated with it:

- Diagnostic toolkits (robopsychology) — explain a specific failure
- Automated evaluators (Azure AI Foundry RAI evals, Promptfoo, DeepEval,
  Ragas, Giskard, Inspect AI, OpenAI Evals) — measure failure rates

The honest framing: the kit is a behavioral intervention. It prevents
at inference time. It does not detect, score, or diagnose. A serious
deployment combines tools from all three rows.

Includes an honest gaps section: no built-in eval harness, fragile
against adversarial users, uneven across languages, doesn't address
agentic loops, punts on the brainstorm-vs-discovery distinction.

deployment-contexts.md covers five concrete patterns with caveats:
personal drop-in, agent runtime skill, CI gate plus evaluator,
layered defense for regulated contexts, research and teaching.
Includes a decision summary table and stackability notes.

README documentation tree and "What it does not do" section now point
at the new docs and at robopsychology for per-incident diagnosis.

No changes to ROOT_PROMPT.md, SKILL.md, the protocol, the essay,
the posts, or the apply_kit.py script.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>