Stabilize randomized bot tests

[8tp]

Summary

• measure bot-emote cooldown from the time an emote is actually emitted, rather than when its randomized delivery is scheduled
• seed the aggressive-versus-conservative challenge-rate simulation so it tests personality behavior deterministically

Root cause

TypeScript 7's dependency PR alternated between green and red without code changes. One failure was a statistical assertion using the process-wide random stream. The other exposed a real timing bug: different randomized emote delays could make delivered reactions land less than the configured cooldown apart even though their scheduling times passed the check.

Impact

CI no longer depends on lucky random samples for this personality assertion, and players receive the full configured cooldown between visible bot emotes.

Validation

• npm run typecheck
• affected 54 tests passed in 10 consecutive runs
• git diff --check
• full unrestricted test/build gate delegated to GitHub Actions because the local managed sandbox blocks the socket E2E suite