Measure the filesystem suite against s3fs on c8gn.2xlarge #1318

Part of #1315. Blocked by #1316 and #1317.

Summary

Run the extended filesystem suite on one c8gn.2xlarge Spot host in us-west-2 and record a history entry comparing S3FileSystem, AioS3FileSystem, and s3fs, with a recommendation for #1315.

Selection

• Single object (Add s3fs as an API of the filesystem benchmark suite #1316): small, medium, large with read (both variants) and write (both variants); xlarge with the in-memory write and copy. Workers 8 and 40 (s3fs max_concurrency), pools 10 and 40, block size 8 MiB, five trials, as in the 4.2.0 run.
• Many objects (Add many-object scenarios to the filesystem benchmark suite #1317): every transport at 1,000 objects (100,000 keys for find), concurrency 8 and 40, five trials.
• The PyAthena cases run in the benchmark environment and the s3fs cases in the s3fs environment of the same checkout; the entry records both environments' library versions.

The host's baseline is 25 Gbps; the 4.2.0 run reached 1,074 MiB/s for 6 GiB writes, so the host is not the limit for the single-object cases.

Deliverables

1. A benchmarks/history/ entry with the environment, selection, run record, findings, limits, and the result tables, following the existing entries.
2. A comment on Benchmark the S3 filesystem against s3fs and decide on a standalone package #1315 with the findings against its expected outcomes and decision criteria, and the improvement issues the results justify (for example parallel batch operations for the synchronous filesystem).

Cost

About 20 to 30 minutes of one Spot host (about $0.10 per hour) and S3 requests below $1; the same order as the 4.2.0 filesystem run.