Docstring-vs-code drift recurs because nothing enforces DIRECTIVE_037 — build the mechanism and update implementer/reviewer guidance

## Problem

A behaviour change lands; the docstring or comment describing the **old** behaviour is never swept. The code is correct and the prose beside it is a lie — which is worse than no prose, because a reader trusts it.

[DIRECTIVE_037 (Living Documentation Sync)](../blob/main/.kittify/) already mandates the behaviour. It is being ignored, and nothing enforces it, so it recurs. Four live instances are linked as sub-issues of this one; each is a one-line fix, and each was found by a squad reviewer reading the code, not by any gate.

The individual fixes are not the cost. **The recurrence is the cost** — this is a defect class that should be closed by construction (DIRECTIVE_043), not point-fixed a fifth time.

## Observed shapes

Worth designing against these specifically, since they are what actually occurs:

1. **A docstring names a pinned version that has since moved.** `zeitgeist_bridge._first_non_printable_attr` cites the `spec-kitty-events` 8.2.0 pin and an encode-side check "in flight" that shipped in 9.1.6 (#4318).
2. **A docstring asserts an exclusion or invariant that a later change abolished.** `registry.py`'s `resolve`/`resolve_local` still describe a routing-catalog project-layer exclusion that #4128 removed; the two catalogs now coincide (#4156).
3. **A docstring overclaims a guarantee the code does not make** (#4038).
4. **A docstring states a contract every caller violates.** `promote_activations`' docstring promises the built-in corpus; every caller passes `default.yaml` (#4400).

Shapes 1 and 4 are mechanically detectable. Shape 2 is detectable when the docstring names a symbol or issue number. Shape 3 generally is not.

## What this issue must deliver

1. **A durable enforcement mechanism.** Design is open — this is deliberately a design issue, not a ready-to-implement one. Candidates worth evaluating rather than assuming:
- a check that any version string appearing in a docstring matches the corresponding pin in `pyproject.toml` / `uv.lock` (closes shape 1 outright);
- a check that symbols named in docstrings still resolve (closes much of shape 2);
- a check that a docstring citing a closed issue number is reviewed when that issue closes;
- failing all of those, an explicit reviewer-checklist gate rather than an implicit directive.
2. **Implementer and reviewer guidance updates.** DIRECTIVE_037 exists and is not landing. Whatever changes — the directive's wording, the review contract, the PR template, the squad prompt — should make "did this change invalidate prose next to it?" an actual step rather than an aspiration.

## Scope note

The four linked sub-issues can be fixed independently and immediately; they do not block this. Closing them without landing the mechanism leaves the class open, which is precisely the pattern this issue exists to break.

Start by reading DIRECTIVE_037 under .kittify/ and the four observed drift shapes, then inspect the relevant version pins in pyproject.toml and uv.lock. Evaluate the proposed enforcement approaches and define a mechanism plus implementer/reviewer guidance that makes prose-impact review an explicit step and covers the feasible cases.

Written by the indexing model from the issue text.