Fix asyncio.get_event_loop() deprecation for Python 3.10+

[waybarrios]

Summary

Fixes #1

• Replace deprecated asyncio.get_event_loop() with asyncio.new_event_loop() + asyncio.set_event_loop() for Python 3.10+ compatibility
• This resolves the issue when running vllm-mlx serve in simple mode

Changes

• vllm_mlx/server.py: Use new_event_loop() pattern instead of deprecated get_event_loop()

Test

vllm-mlx serve mlx-community/Llama-3.2-3B-Instruct-4bit --port 8000

Thanks @rhychung for reporting and suggesting the fix!

[waybarrios]

Benchmark Results

Tested with the model from the issue report:

============================================================
BENCHMARK RESULTS
============================================================

Model          mlx-community/Llama-3.2-3B-Instruct-4bit
Hardware       M4 Max (128 GB)
Total Runs     3
Input Tokens   21
Output Tokens  283
Total Time     1.58s

Performance Metrics:
Metric                        Mean         P95/Max
----------------------------  -----------  -----------
TTFT (Time to First Token)    58.9 ms      62.1 ms
TPOT (Time Per Output Token)  4.99 ms      5.01 ms
Generation Speed              200.3 tok/s  200.6 tok/s
Processing Speed              118.5 tok/s  -
Latency (per request)         0.52s        0.55s

Throughput:
Total Throughput  192.9 tok/s
Requests/Second   1.90 req/s

Resource Usage:
Process Memory (peak)  2.43 GB
MLX Peak Memory        1.72 GB
MLX Cache Memory       0.03 GB
System Memory          41.3 / 128 GB (32%)
============================================================

All tests pass and benchmark runs successfully with the fix.