docs: add deployment-contexts; expand related-work with public RAI evaluator landscape

New deployment-contexts.md addresses the most common question I get when
introducing the toolkit: "we already run automated evaluators in CI, why
also use this?". It frames three classes of tools — automated evaluators,
diagnostic toolkits, behavioral interventions — and shows where
robopsychology fits, when it is the wrong tool, and three practical
composition patterns including a layered defense for regulated contexts.

Expanded related-work.md with two new sections:

- Automated RAI / behavioral evaluators (2024-2026 landscape): public
  evaluator stacks that have matured since v2.5 — Azure AI Foundry RAI
  evaluators, Promptfoo, DeepEval, Ragas, Giskard, OpenAI Evals — plus
  the explicit framing that they are complements (quantification across
  batches), not substitutes (causal explanation per case).
- Behavioral interventions (the third class): system prompts,
  constitutional AI, RAG with citation enforcement, RLHF tuning, and
  baloney-detection-kit as a sister project on the prevention side.

README documentation index now links deployment-contexts.md.

No code changes. No changes to existing prompts, prompts.yaml, CLI, or
validation cases.

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