fix(CLAUDE.md): .hermit-quit now cancels the recurring cron trigger, not just the current pass

[ytnobody]

Summary

Closes #192.

Superintendent cycle step 2 said .hermit-quit is "a terminal stop" that
requires starting /hermit again to resume — but in practice it only
skipped work for the pass that detected the file. Any recurring
session-only cron job registered in step 1 kept firing /hermit every
tick regardless, so every subsequent tick burned a full pass just to
re-detect the same quit file, until the Claude session itself expired
(up to 7 days) or someone manually ran CronDelete.

• Updated CLAUDE.md's Superintendent cycle step 2 to explicitly call
  CronList to find the recurring /hermit job and CronDelete to
  cancel it before ending the pass, so quit actually stops the
  autonomous loop rather than just leaving it spinning on no-ops.
• Investigated hermit run (Issue hermit run: ループを Claude Code セッションの外に出し、HERMIT 自身のプロセスが所有する #181 / PR feat(run): add hermit run — Superintendent loop owned by HERMIT's own process #189, internal/runloop/runloop.go):
  it's a separate OS process loop that already watches .hermit-quit
  directly (runloop.go:145) and stops its own tick loop — it never
  registers or depends on session-only CronCreate/CronList/CronDelete
  jobs, so it isn't affected by this bug. Noted this explicitly in the
  step 2 text so the distinction is documented rather than assumed.
• Applied the identical fix to cmd/hermit/templates/CLAUDE.md.tmpl
  (the scaffold used by hermit init) to keep the two in sync, per
  existing repo convention.
• Left step numbering, .hermit-paused handling (step 3), and step 1's
  cron-registration logic untouched — this is purely additive detail
  inside step 2, no steps inserted/removed/renumbered.

Assumptions

• Kept the added detail as sub-bullets under step 2 rather than
  inserting new top-level numbered steps, to avoid renumbering every
  step that's cross-referenced by number elsewhere in the document
  (e.g. step 5/6/7's back-references). This satisfies the Issue's
  scope constraint ("changed only step 2's procedure").
• No Go test exists asserting on CLAUDE.md's Superintendent-cycle step
  content/structure beyond substring checks in
  cmd/hermit/inprocess_test.go (e.g. presence of
  ### Superintendent cycle (one pass, run inline), which still pass;
  no new test was added since this is a documentation-only change with
  no new tool-calling behavior to unit test.

Test plan

• go build ./...
• go test ./cmd/hermit/... ./internal/runloop/... (all pass)
• Manually diffed CLAUDE.md vs cmd/hermit/templates/CLAUDE.md.tmpl to confirm step 2 text stayed in sync

[ytnobody]

Superintendent review (HIGH risk — CLAUDE.mdがhigh-riskパスのため)

diffを確認しました。CLAUDE.mdとその複製であるcmd/hermit/templates/CLAUDE.md.tmplのstep 2記述に、.hermit-quit検知時にCronList/CronDeleteで既存の定期トリガーを実際にキャンセルする手順を追加しただけの、純粋なドキュメント修正です（2ファイル、+8/-2行）。

• hermit run（OSプロセスループ）は.hermit-quitを直接監視しており本問題の影響を受けないことをコード（internal/runloop/runloop.go）から確認した旨も明記されており、Issue #192の要件・受け入れ条件を満たしています。
• 両ファイルの記述は同期しており、既存のステップ番号への影響もありません。
• CIは全てpass、developとのマージ状態もCLEANです。

安全にマージできると判断します。

[ytnobody]

⚠️ HERMIT: HIGH risk detected.
Reasons: [CLAUDE.md is in a high-risk path]