feat(observability): logging standard covers every process type, not just API calls

[nfeuer]

Summary

The kit's logging guidance skewed LLM/API-call-centric: the only itemized field requirements anywhere were tokens/cost/task_type, CLAUDE.template.md's logging principle specified only model calls, and nothing stated what a background job, queue consumer, batch stage, state machine, or the process lifecycle must emit. This PR makes the logging instructions complete and useful for every process type:

• docs/LOGGING_STANDARD.md (new) — per-process-type event catalog with minimum required fields: requests, external calls (LLM as one case), background jobs (start/progress/terminal with items processed/skipped/failed, overlap detection), queue consumers (per-message outcome, retries, dead-lettering), batch/ETL stages (rows in/out, per-record skip reasons), process lifecycle (startup/shutdown, supervised task exits), state transitions, retries/fallbacks, caches/pools — plus the anti-patterns (failure-only logging, start-without-end, counts-without-reasons, prose events).
• templates/logging_setup.py — new log_operation context manager: paired <name>_completed/<name>_failed events with duration_ms for any operation, re-raising on error. Makes the universal rule ("every significant operation logs outcome + duration") a one-liner.
• observability-check skill — telemetry step broadened beyond external calls; new long-running-process legibility step; job/run IDs treated as correlation context; missing-heartbeat failure mode (absence of job_completed is the alert); report block extended.
• observability-reviewer agent — reads the catalog first; new job/consumer/pipeline lifecycle category; flags success-silent operations (indistinguishable from never-ran).
• CLAUDE.template.md principle 3 generalized — model-call fields become the additional LLM requirement, not the whole rule.
• Wiring — bootstrap installs LOGGING_STANDARD.md into new projects; README tree/catalog updated; kit_version 1.1.0 → 1.2.0 so kit-update offers this to adopted projects.

Spec citation

• No design change — the kit repo has no canonical spec; the kit's own convention docs (LOGGING_STANDARD.md) are the deliverable.

What was tested

• Ran kit-ci's validation locally: templates/logging_setup.py compiles (py_compile), kit.yaml parses, changed skill/agent files have valid frontmatter.
• Unit tests: N/A — documentation/template repo.
• Manual verification: cross-references resolve — LOGGING_STANDARD.md is referenced from observability-check, observability-reviewer, CLAUDE.template.md, logging_setup.py, and the bootstrap install step; reviewer's category count updated to match its new section.
• Coverage ratchet: N/A

Risk and rollback

Risk: low — documentation, a new additive template helper, and checklist wording; no hooks or gate semantics change. Adopted projects see nothing until they run kit-update.

Rollback: revert the PR.

Docs

• Narrative docs updated in the same PR (LOGGING_STANDARD.md, README, CLAUDE.template.md)
• Changelog entry — generated at next release by the release skill
• No ADR needed — convention documented in LOGGING_STANDARD.md itself

Follow-ups

• None added or closed

---

🤖 Generated with Claude Code

https://claude.ai/code/session_01LxEwyqF1KjEPpQKqM5tSrv

---

Generated by Claude Code