fix(ch06): improve MC distractor quality for quiz questions

[calmcoconut]

Summary

Audited all multiple choice questions in Chapter 6 and replaced weak distractors that were too easily eliminated without knowing the material. The goal is that every answer choice should be plausible enough to require genuine understanding to rule out.

Changes

Q2 — Synchronous Replication Trade-offs

• Option D (replaced): Previous distractor referenced "rolling schema migrations / network jitter" which has no connection to synchronous replication. Replaced with an accurate-sounding description of sync replication that forces students to reason carefully between all four options.

Q4 — Purpose of Taking a Snapshot

• Option C (replaced): "Run checksum validation on block files" is a real DB operation but obviously irrelevant to follower setup. Replaced with a common misconception DDIA explicitly corrects: that you need to lock leader tables during snapshot creation.
• Option D (replaced): "Run a vacuum to reclaim disk space" similarly irrelevant. Replaced with a realistic conflation: capturing the LSN/binlog position (a real step) vs. the snapshot itself.

Q6 — Split-Brain Definition

• Options A, C & D (replaced): All three were hardware-level failure modes unrelated to distributed systems. Replaced with:
  • A follower serving stale reads due to replication lag (confuses split-brain with lag)
  • Consensus failing to elect any leader — the opposite failure mode
  • A fenced old leader still receiving writes — a real edge case students mix up with split-brain

Q17 — Leaderless Write Propagation

• Option C (replaced): "Replica nodes poll client-side memory buffers" is too absurd to be a credible distractor. Replaced with a gossip-protocol trap: students who know Cassandra uses gossip for metadata propagation may incorrectly apply it to data write propagation.

Questions Left Unchanged

Q1, Q3, Q5, Q7, Q8, Q9, Q10, Q11, Q12, Q13, Q14, Q15, Q16, Q18 — all deemed well-constructed with plausible distractors.