perf(rustfs): remove the fake s3_operations benchmark (#4817)

bench(rustfs): remove the fake s3_operations benchmark

rustfs/benches/s3_operations.rs never measured S3: all three groups
(put_object/get_object/list_objects) only black_box(data.len())/black_box(count)
with a 'In a real benchmark, this would call the actual S3 client' comment. It
gave a false 'S3 operations have benchmark coverage' signal and was dead weight
on rustfs/Cargo.toml.

Delete the file and its [[bench]] entry. S3-face performance is covered solely
by the warp A/B gate (performance-ab.yml); the runbook scope note now says so
explicitly. Micro-benchmarks stay at the function level (perf-8).

Refs rustfs/backlog#1152 (perf-9).

docs/operations/hotpath-warp-ab-runbook.md

@@ -155,3 +155,10 @@ The gate logic is unit-validated across pass/warn/fail/exempt outcomes; the
 orchestrator and workflow are shellcheck- and `--dry-run`-validated. The first
 real warp measurement belongs on a Linux runner or the ansible cluster — there
 is no warp/multi-disk rig in the repo's local checkout.
+
+This warp A/B gate is the **only** entry point for S3-face (PutObject /
+GetObject / ListObjects) performance coverage. There is deliberately no
+in-process criterion benchmark for those operations: a criterion harness that
+stands up an embedded server measures the harness, not the S3 path, so it would
+report a number without guarding anything. Micro-benchmarks stay at the
+function level (EC encode, `xl.meta` parse, `rename_data`; perf-8).