[Enhancement]: Experiment with a different WSGI HTTP server

### Affected Area

Performance

### Problem Statement

Experiment with a slightly better!? web server.

Some eyeballed/unreproduced tests on my production bench.

```bash
granian --host 0.0.0.0 --port 8000 --interface wsgi --workers 3 --respawn-failed-workers --workers-kill-timeout 30 --no-reload frappe.app:application
```
instead of
```bash
gunicorn --bind 0.0.0.0:8000 --workers 3 --timeout 120 --graceful-timeout 30 --worker-tmp-dir /dev/shm frappe.app:application --preload --statsd-host=10.25.9.108:9125 --statsd-prefix=bench-31400-000001-u8-nuremberg --worker-class=gthread --threads=4 --max-requests 5000 --max-requests-jitter 1000 --access-logfile /home/frappe/frappe-bench/logs/u8-nuremberg.frappe.cloud-gunicorn.access.log
```

https://github.com/frappe/agent/blob/522fdf00e4f61dc0d14aebec3f35ffe0feb2952f/agent/templates/bench/supervisor.conf#L8

Better RSS. Better SHR. Needs confirmation.
After

Before

### Proposed Solution

_No response_

Start with the referenced agent/templates/bench/supervisor.conf line 8 and compare its Gunicorn setup with the provided Granian command. Run reproducible benchmarks that measure RSS and SHR, then document whether the alternative is safe to adopt and what configuration changes are required.

Written by the indexing model from the issue text.