Increase Modal throughput with per-container input concurrency

Documented steps

1. Identify whether the workload is I/O-bound (external API calls, database waits) or batchable (GPU inference) — those are the cases where one container should handle several inputs at once.
2. Add the decorator beneath the Function decorator: `@app.function()` then `@modal.concurrent(max_inputs=100)`.
3. Apply the same decorator to class methods under `@app.cls()` so a single loaded model serves concurrent inputs.
4. Make the function body genuinely concurrent — use async/await or thread-safe client libraries. A synchronous CPU-bound body will not benefit and will just queue inside the container.
5. Size max_inputs against the container's memory: N concurrent inputs means N simultaneous copies of per-request state.
6. Re-tune max_containers after enabling concurrency; the container count needed for the same throughput drops, so a previously correct ceiling may now be far too high.
7. Reference: https://modal.com/docs/guide/webhooks