perf: switch queue to use a ring buffer instead of a linked list

Initially, we cared a lot about simplicity and correctness. Now, grammY powers some of the largest bots on Telegram, so we should start optimizing down to the last allocation.

Linked lists likely cause a lot more memory overhead (even in V8) and we should:
1. implement a benchmark for throughput
2. implement metrics for memory churn, CPU performance, and other relevant factors
3. implement a ring buffer version
4. run the benchmarks, and compare the implementations
5. switch if the results indicate this