Add patterns 34-35: uniform sentence and paragraph length

[MCKRUZ]

What this adds

Two patterns covering the structural layer: #34 Uniform Sentence Length and #35 Uniform Paragraph Length.

Patterns 1-33 operate at the word and phrase level. Prose can satisfy all of them and still read as generated, because LLMs settle into a narrow sentence-length band and stay there, and produce paragraphs of even size and identical internal shape.

Why it isn't already covered

DETECTION GUIDANCE already lists "Variety in sentence length" under Signs of human writing. That is a detection signal, telling the editor what not to destroy. There was no generative counterpart telling the editor how to fix prose that lacks it.

These two patterns supply the method: count words per sentence across a few paragraphs, and if the spread is narrow, break one sentence hard and let another run long.

Guards against fighting existing patterns

I was conscious that a naive "vary your sentence length" rule collides with §31, so:

• Codex support #34 explicitly defers to Skill Expansion: SOTA Tiered Architecture, Governance & Portability #31. Stacking short fragments to manufacture drama is its own tell. The instruction is to increase variance, not to shorten. It also warns against mechanical short-long-short-long alternation, which is just a different uniform pattern.
• Both patterns carry a genre limit. Reference documentation, API docs, procedures, and legal text are uniform by design. A matching entry was added to the false-positive list so low variance in those genres is not flagged.

Housekeeping, per AGENTS.md

• README pattern table and "N Patterns Detected" heading updated to 35
• AGENTS.md pattern count updated
• metadata.version and .claude-plugin/plugin.json bumped to 2.10.0 (kept in sync, the validator checks this)
• scripts/validate-package.py bounds updated to 1-35
• README version history entry added
• SKILL.md is 450 lines, within the 500-line portability budget
• python scripts/validate-package.py passes

Unrelated bug you may want to know about

scripts/validate-package.py crashes on Windows with UnicodeDecodeError because read_text() uses the platform default encoding (cp1252) and SKILL.md contains non-Latin-1 characters. This reproduces on unmodified main, so it is not introduced here and I have left it alone. Passing encoding="utf-8" to the three read_text() calls would fix it. Happy to send that as a separate PR if useful.