[tire-fire]

Summary

Start(), waitForReset() and rvb() each built their own Redis client from REDIS_ADDR (falling back to quotient_redis:6379) just to subscribe to the events channel, and rvb() did so every round. They now subscribe through se.RedisClient, the client the engine already uses to publish those events and to move tasks and results. waitForReset becomes a method so it can reach it.

This also fixes the engine tests. They publish to the test Redis (REDIS_HOST/REDIS_PORT, port 6380 in CI), but rvb() subscribed wherever REDIS_ADDR pointed, and two tests set that to localhost:6379. So in tests the engine never saw events published during a round. Those two t.Setenv("REDIS_ADDR", ...) calls are removed, since nothing reads it on that path anymore.

The runner still reads REDIS_ADDR for its own connection; that's unchanged.

Tests

• New TestRvb_StopsOnResetPublishedToEngineRedis publishes reset on the engine's Redis client during result collection and expects rvb() to stop with "reset event received". It fails against the current engine.go (rvb() returns nil after the round times out) and passes with this change.
• The full suite, go vet and golangci-lint pass.