Run loadtest with tabby on modal GPUs

Steps:

1. Adjust TABBY_API_HOST in run.sh to match your modal deployment url.
2. Add models you're interested in to benchmark at end of run.sh
3. Run run.sh, output will be appended to record.csv