fix(v3): adopt v4 dequeue strategy to increase perf

We will now dequeue into a redis list as many messages as we possibly 
can from actual queues, trying to dequeue up to 10 runs at a time for 
each queue in an env, in separate queue consumers. Then the 
SharedQueueConsumer can simply `lpop` off that list and process a run to
be dequeued very quickly. This should get runs through the platform side
dequeue system very quickly.

Author: ericallam

apps/webapp/app/env.server.ts

@@ -410,6 +410,8 @@ const EnvironmentSchema = z.object({
   MARQS_QUEUE_AGE_RANDOMIZATION_BIAS: z.coerce.number().default(0.25),
   MARQS_REUSE_SNAPSHOT_COUNT: z.coerce.number().int().default(0),
   MARQS_MAXIMUM_ENV_COUNT: z.coerce.number().int().optional(),
+  MARQS_SHARED_WORKER_QUEUE_CONSUMER_INTERVAL_MS: z.coerce.number().int().default(250),
+  MARQS_SHARED_WORKER_QUEUE_MAX_MESSAGE_COUNT: z.coerce.number().int().default(10),
 
   PROD_TASK_HEARTBEAT_INTERVAL_MS: z.coerce.number().int().optional(),