feat: add prose-token-leak lint rule

[zachshallbetter]

Added a new lint rule, prose-token-leak, to flag literal token value duplication in markdown body text to prevent documentation drift (resolves Issue #16).

Background & Motivation

The spec explicity separates machine-readable tokens (YAML frontmatter) and human-readable context (markdown body). When authors duplicate literal values (such as hex colors #ffffff, functional colors rgb(...) / hsl(...), or dimensions 16px) in their prose descriptions or tables, it introduces a dual source of truth. If token values are updated, the prose is frequently forgotten and drifts out of sync.

As discussed in Issue #16, there is community consensus that:

• The markdown body should describe design intent and perceptual registers, but never contain literal values.
• Prose should instead reference tokens by name (e.g., {colors.primary}, body-md type scale).
• A linter warning is the natural guardrail to catch these leaks during development before they are committed.

This rule builds on the pattern established in PR #155 (custom linter checks and omitted-sections validations) by keeping standard linter warnings as the primary gate for format compliance.

Relation to other PRs

• PR feat: add omitted frontmatter key to suppress expected-missing lint warnings #155 (Merged): This continues validation work on linter rules (which added the omitted validations in PR feat: add omitted frontmatter key to suppress expected-missing lint warnings #155).
• PR feat(spec): add 9 standard CSS component sub-tokens #89 (Closed): PR feat(spec): add 9 standard CSS component sub-tokens #89 proposed expanding the schema vocabulary to cover a large list of CSS properties. The maintainers closed it to keep the schema a "structural minimum" rather than chasing full CSS property coverage. This prose-token-leak rule directly reinforces that philosophy by steering authors away from writing hardcoded literal values in prose and encouraging clean token usage.
• PR feat: support axis-specific padding component sub-tokens #166 (Pending): This contribution is completely independent of the pending axis-padding PR.

Changes

• Model State:
  • Extend DesignSystemState to include documentSections to expose parsed section content to the rules.
  • Forward input.documentSections in ModelHandler.
• Linter Rule & Registration:
  • Add prose-token-leak.ts containing the validation logic. Regex patterns flag literal hex colors, functional color notations (rgb, rgba, hsl, hsla), and CSS dimensions.
  • To prevent false-positives, the rule strips out:
    • Markdown links and HTML tags (ignoring anchor targets).
    • HTML comments, headings, fenced code blocks, and inline code backticks.
    • Curly-brace token references (e.g., {colors.primary}).
    • The YAML frontmatter block inside the prelude.
  • Register the rule in rules/index.ts and update rules count assertions in test files.
• Documentation:
  • Update spec.mdx to define the prose token duplication policy.
  • Regenerate docs/spec.md (see docs/spec.md#L8).

[zachshallbetter]

Also, future extensions might/could include:

• Font Families: Warn on registered font-family names written as raw text in prose (forcing references like {typography.body.fontFamily}).
• Font Weights: Flag raw numeric weights (e.g., 600, 700) when close to keywords like weight or fontWeight.
• Box-Shadows: Match literal CSS box-shadow styles once elevation/shadow tokens are standard in the spec.
• Custom Token Keys: Warn on raw custom token keys in prose unless formatted as code ticks (e.g., `brand-primary`) or references.
• Breakpoints: Flag hardcoded pixel/em media query widths (e.g., @media (max-width: 1024px)) once breakpoint tokens are supported.
• Z-Index Layering: Flag literal z-index declarations in prose.