Enforce docstring-presence rules incrementally (D100–D107)

[dgarros]

Why

Follow-up to #1184. The bulk of the docstring backlog is "missing docstring"
(presence) rules — ~2,275 violations. Unlike the consistency rules, these require
writing docstrings, so a big-bang is impractical. This issue tracks clearing
them incrementally, public API first, without blocking the whole rule set.

Goal: enforce presence rules on the public surface first, then peel off the rest,
so the parked list becomes a visible, shrinking to-do map.

Non-goals: the consistency rules (D417/DOC502/DOC201) — tracked in the Phase 1
step 2 issue.

Scope (currently parked)

Code	Rule	Count
D102	undocumented-public-method	705
D103	undocumented-public-function	632
D101	undocumented-public-class	536
D100	undocumented-public-module	217
D107	undocumented-public-init	91
D104	undocumented-public-package	55
D105	undocumented-magic-method	39

Total ≈ 2,275 violations.

Approach

Convert the global park into shrinking per-file-ignores rather than a single
global ignore:

1. Enforce D101/D102/D103 on the public surface first — client.py, node/,
   config.py — and keep them parked only for tests/, ctl/, and internal modules.
2. Peel off directories as they get documented; each PR shrinks the per-file-ignore list.
3. Document lower-priority areas (magic methods D105, __init__ D107, package
   docstrings D104) last.

Consider splitting into per-area sub-issues once the public-API surface is done.

Done when

• D100–D107 are enforced repo-wide (no remaining per-file-ignores for them).
• uv run ruff check is green.

Related

• Ratchet idea (separate future issue): a CI step running ruff check --select D,DOC
  on the diff only, so new code must be clean while the historical backlog is parked.