#105 fix(scaling-dive): the 250-author cliff was rate-limit colocation, not a real ceiling

Raise commitRateLimiting points ceiling to 1,000,000 in the dive workflow so colocation of harness + SUT on one runner stops triggering it. NODE_ENV=production enables Etherpad's per-IP rate limit; handleMessage applies it to every msg including CLIENT_READY; all simulated authors share 127.0.0.1 = one bucket = trips at N * editRate. Real deployments with many client IPs are not affected.