docs: rewrite safety/errors/playbooks/skills/observability/server/rag/checkpointers

[fede-kamel]

Summary

Third concept-page batch. Same template as PRs #29 and #30: pitch /
when to pick / getting started / capabilities (each runnable) /
gotchas / source / see also.

Notable: several pages were aligned with the real API after I
read the source. The previous copy described surfaces that don't
exist:

• safety.md (81→165) — `GuardrailsHook(config=GuardrailConfig(...))` with the actual fields (block_dangerous_tools / pii_patterns / blocked_content_patterns / max_prompt_length / default_action). Previous version invented a `TopicPolicy(deny=...)` API.
• errors.md (60→127) — full `LocusError` hierarchy with stable `kind` strings; idiomatic one-handler / kind-keyed metrics / differentiated retry / chained-cause patterns.
• playbooks.md (72→178) — real `PlaybookStep` schema (`id / description / expected_tools / hints / required / max_tool_calls / validation`) + `PlaybookEnforcerHook`. Previous version used a fictional `action / args / expect / requires` schema.
• skills.md (70→163) — real `Skill` class + AgentSkills.io progressive-disclosure story (catalog → instructions → resources), `allowed_tools` scoping, Skill vs Playbook vs Tool disambiguation.
• observability.md (72→166) — actual metric names (`locus.invocations`, `locus.iterations`, `locus.tool_calls`, `locus.tool_errors`); PII guidance for `record_arguments` / `record_results`.
• server.md (67→178) — real `AgentServer` auth model (bearer `api_key`, loopback fallback, per-principal thread scoping that prevents cross-client thread guessing).
• rag.md (83→229) — `create_rag_tool(retriever)` factory (previous version called `retriever.as_tool()` which doesn't exist), embedder/store matrices, hybrid retrieval guidance.
• checkpointers.md (96→181) — clarifies the two checkpointer shapes: native `BaseCheckpointer` vs storage backends needing the `*_checkpointer()` factory adapter.

Test plan

• `hatch -e docs run mkdocs build --strict` — clean (0 warnings, 0 broken anchors)
• `pre-commit run --files docs/concepts/{safety,errors,playbooks,skills,observability,server,rag,checkpointers}.md` — codespell + markdownlint pass
• DCO sign-off on the commit