F6: Add per-namespace blurbs to `dialect.yaml` so agents can disambiguate borderline keys

[ChauCM]

Surfaced by a pilot test (Claude Code Opus 4.7 against examples/after/, 2026-05-24). Full feedback: dialect-feed.md.

Problem

Today the meaning of each namespace lives only in @key.description ad-hoc. An agent classifying a new key has to read every existing key in that namespace to understand the namespace's purpose.

Fix

Add a namespaces: block in dialect.yaml where each entry can include a one-liner:

namespaces:
  checkout: "Booking confirmation flow — payment + agreement screen."
  settings: "User account preferences."
  home: "Trip list, top-level navigation."
  common: "Shared UI strings (cancel, save, loading)."

The agent reads these blurbs at extract time to decide where a new key belongs ("does this go in common or settings?") without having to scan every existing key.

Acceptance

• dialect.yaml supports a namespaces: block with per-namespace descriptions.
• Init plan / describe plan templates teach the agent to read these blurbs when deciding namespace placement.
• dialect check flags namespaces that lack a blurb (warning, not error).