Build load tests for RPC endpoints

[TomasArrachea]

Follow-up issue for #609.

Use the loaded store generated with the seed-store CLI added in #657 and collect performance metrics when executing RPC endpoint calls.

[TomasArrachea]

#661 introduces an initial approach to benchmark the sync_state request. Currently, it executes multiple requests sequentially and prints the average elapsed time per request. I think it's worth discussing which metrics are most valuable for this flow (and also others as well).

For example, instead of only sequential execution, we could run the requests at different levels of concurrency and measure measure latency percentiles (e.g. p95, p99).

---

In #778 there is a possible implementation on running the requests concurrently and measuring the p95 stat.

[bobbinth]

For example, instead of only sequential execution, we could run the requests at different levels of concurrency and measure measure latency percentiles (e.g. p95, p99).

I agree that running at different levels of concurrency is a good idea. I think we should also run at different database sizes - e.g., 100K vs. 1M (ideally, we'd do bigger database sizes too - but that could be further down the road).

As for latency percentiles, I think p95 is probably fine.