docs(hooks): narrow denylist header comment to its actual scope (#314)

[dngioidev]

What & why

The plugin/hooks/denylist.mjs header comment overstated coverage — "Blocks
destructive commands" reads as a general destructive-command sandbox. In reality
the hook is a targeted backstop over a fixed pattern set (git history/branch
operations, rm recursive-force outside build/temp dirs, and pipe-to-shell /
eval-of-substitution RCE) that FAILS OPEN. This rewrites the header doc
comment to state the real scope and the fail-open stance, so it is not mistaken
for a security boundary.

Documentation-accuracy fix only. No rule logic or runtime behavior changed —
only the header comment block was edited.

Closes #314

Acceptance criteria

• AC.1 — the denylist.mjs header comment accurately states it is a
  TARGETED backstop (git history/branch/reset/clean + rm recursive-force +
  pipe-to-shell/eval) that FAILS OPEN, not a general destructive-command sandbox.
  Verified: new header enumerates the exact RULES set, explicitly says "NOT a
  general destructive-command sandbox" and "FAILS OPEN by design ... exit 0 ...
  can never take the session down"; behavior unchanged.

Verification

• pnpm verify (vitest) — 54 files / 618 tests, all green (comment-only change,
  no test drift).
• claude plugin validate ./plugin --strict — validation passed.
• Diff is a single 12-insertion/3-deletion comment block; git diff confirms no
  code lines touched.
• No new test added: this is a doc-accuracy AC; the CI verify job runs vitest run only (acgate is not a CI gate here and there is no plan file mapping
  AC-314.1), so scope was kept minimal per ticket guidance.