Combinatorics bundle + fix for daily-run duplication (per-subtopic numbering & coordination protocol)

[DMDaudio]

What does this PR add or change?

Two things: a new content bundle, and a fix for the process problem that has been generating
overlapping, colliding daily-run PRs.

1. New content — Quant counting-probability / combinatorics

First real coverage of combinatorics (only a single probability question existed before):

• 1 lesson (quant-counting-probability-combinatorics) — fundamental counting principle,
  factorials, permutations vs. combinations (C(n,k) = P(n,k)/k!), identical-item arrangements,
  and together/apart restrictions.
• 7 problem-solving questions (0001–0007) with an easy→hard progression: counting principle
  → 5! → C(8,3) → LEVEL anagrams → glue method → "at least one" via complement → "no two
  together" via gap method.
• Distractors keyed to specific errors; worked explanation + two hints each; every answer verified.

2. Fix — stop daily runs duplicating each other

Root cause: daily-run agents each branch off main and never merge, so a run only sees main
and is blind to other in-flight runs. Combined with a section-wide NNNN counter, every run
grabbed the same "next" number and re-seeded the same subtopics — producing ~15 overlapping open
PRs that collide on both subtopic and numbering.

• Per-subtopic numbering. NNNN is now unique within the subtopic (new subtopics start at
  0001), not section-wide. Two branches now only collide if they target the same subtopic —
  which the protocol forbids. Documented in CONTENT_SCHEMA.md and CONTRIBUTING.md. No existing
  files need renaming (they're already unique within their subtopic). This branch's combinatorics
  questions were renumbered 0017–0023 → 0001–0007 to follow the new scheme.
• Coordination protocol (new section in CONTRIBUTING.md): before claiming a subtopic, check
  the repo's open pull requests (not just main); one subtopic per run; prefer genuine gaps;
  keep each run on its own PR.
• scripts/coverage.mjs — dependency-free report of per-subtopic question counts (by
  difficulty), lesson presence, and gaps from the current checkout; also flags content tagged
  outside the taxonomy. It states up front that it only sees the checkout, so runs must still check
  open PRs.

Limitation: this fixes the go-forward process. The existing open-PR backlog still needs the
maintainer to merge or close it to clear the overlaps already created.

• Rebuilt docs/data/content.json via node scripts/build.mjs (39 questions, 9 lessons; zero
  warnings).

Type of change

• New question(s)
• New lesson(s)
• Fix to existing content (answer / explanation / tag)
• Tooling / docs / schema (numbering convention, contributor protocol, coverage script)

Author checklist

• This is my own original work — not copied or paraphrased from any real GMAT question or prep material.
• Front-matter follows CONTENT_SCHEMA.md (id matches the filename; topic/subtopic exist in curriculum.md).
• I solved every question myself and confirmed exactly one answer choice is correct.
• Each explanation would help someone who got the question wrong, and names the trap where relevant.
• status is in-review. I did not set my own content to verified.

For reviewers

• I solved the question(s) without reading the keyed answer first and got the same answer.
• Exactly one choice is correct; distractors are plausible.
• Explanation and tags (section / topic / difficulty) are accurate.
• The per-subtopic numbering change and coordination protocol read sensibly for future runs.
• If approving: bumped status to verified and added my handle to reviewers.