safekeeper: add WAL ingestion benchmarks with pageserver fan-out

[erikgrinaker]

For sharded ingest performance, we should add a benchmark that ingests WAL into a Safekeeper and fans it out into 8 shards across 8 pageservers.

Ideally, we would have a full end-to-end benchmark including compute, safekeeper cluster, and pageserver cluster with networking and S3. However, building this infrastructure is likely more work than we have resources for in this project.

For now, we'll write a Rust integration-style benchmark that runs the Safekeeper and Pageservers in memory on the local node, ignoring network latency, (possibly) disk latency, and compute overhead.