ci(perf): fix nightly warp A/B startup failure, make it diagnosable, add small-object + 10MiB cells (#4669)

Both nightly runs of the warp A/B rig failed at server bring-up:
`endpoint http://127.0.0.1:9000/health did not become healthy` — exactly 60s
after start. The server binds its public listener only after startup converges
(erasure-format load + IAM); its own budget
(RUSTFS_STARTUP_READINESS_MAX_WAIT_SECS) defaults to 120s, so the rig's 60s
health poll gave up before a slow cold start on the shared sm-standard-2 runner
finished. The rustfs.log lived under /tmp (not the uploaded target/hotpath-ab/),
so the root cause was invisible in the artifact.

Root-cause + diagnosability (scripts/run_hotpath_warp_ab.sh):
- Health poll is configurable via --health-timeout, default 180s (> the 120s
  server startup budget). Fails fast if the local server process exits before
  becoming healthy instead of polling out the full window.
- Server logs + a startup-env file now write under OUT_DIR/server-logs/ so they
  ride along in the CI artifact. On a health failure the rig dumps the last 50
  log lines to the job log.

Workflow (.github/workflows/performance-ab.yml):
- On every run, write gate.md (or, on a pre-gate failure, the failing phase's
  server-log tails) into $GITHUB_STEP_SUMMARY; short artifact retention.
- Short measurement matrix (--duration/--rounds/--cooldown) so the expanded
  matrix fits; timeout-minutes 90 -> 120 as a Phase-0 stopgap until perf-3
  caches the baseline binary (the ~65min double build dominates).
- TODO(ci-8/perf-2) hook comment for the failure-alert composite action.

Workload cells (absorbed perf-4): add put-4kib/get-4kib (the #4221 fsync
regression size, ~-10% @4KiB, previously invisible) and get-10mib (historical
large-GET EOF size) to both the PR-labeled and nightly matrices — 6 workloads
x 2 phases x 2 drive-sync = 24 cells. A 1KiB cell is deferred to protect the
budget until perf-3.

Refs rustfs/backlog#1152 (perf-1, absorbed perf-4), rustfs/backlog#1155.

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

@@ -23,6 +23,29 @@ Metric directions: `reqps` (put obj/s) and `throughput` (get MiB/s) are
 higher-is-better; `latency` / p99 (mixed) is lower-is-better. warp is assumed
 pre-installed, as elsewhere in `scripts/`.
 
+## Workload matrix
+
+Six workloads × the drive-sync on/off matrix × baseline/candidate = 24 cells:
+
+| Workload | mode | size | why |
+| --- | --- | --- | --- |
+| `put-4kib` / `get-4kib` | put / get | 4KiB | the #4221 fsync regression size (~-10% @4KiB) — previously invisible |
+| `put-4mib` / `get-4mib` | put / get | 4MiB | bulk obj/s and MiB/s |
+| `get-10mib` | get | 10MiB | the historical large-GET EOF size |
+| `mixed-256k` | mixed | 256KiB | p99 latency |
+
+Sizes are passed to the load driver via `--sizes` (one size per cell); the
+driver's `DEFAULT_SIZES` covers 1KiB..10MiB, so any of those can be added by
+editing `WORKLOADS` in `scripts/run_hotpath_warp_ab.sh`. A 1KiB cell is left
+out for now to keep the nightly matrix under the 90-minute job budget (the
+baseline+candidate double build dominates until perf-3 caches it); re-enable it
+once that lands.
+
+CI runs a **short** warp matrix (`--duration`/`--rounds`/`--cooldown` tuned in
+`.github/workflows/performance-ab.yml`) so all 24 cells fit the budget without
+dropping cells. These params are deliberately noisy-but-fast for the Phase-0
+"keep the pipeline alive" goal; perf-6 recalibrates them.
+
 ## Local mode (quick / CI smoke)
 
 Builds both binaries and runs a throwaway single-node server on local disks.
@@ -94,6 +117,22 @@ cost to restore power-loss durability. For those, run with
 `--allow-regression` (or add the `perf-deliberate-tradeoff` label in CI): the
 FAIL is recorded and rendered as an exempted WARN, and the gate exits 0.
 
+## Diagnosing a failed run
+
+Each phase's server log and its startup environment are written under the run's
+output dir (`target/hotpath-ab/<ts>/server-logs/<phase>-sync-<sync>.{log,env}`)
+and uploaded in the `hotpath-warp-ab-<run>` artifact, so a failure is
+diagnosable after the fact. On a health-check failure the rig also dumps the
+last 50 log lines into the job log and the CI job writes the failing phase (or
+the gate table) into the GitHub step summary.
+
+Readiness polling waits up to `--health-timeout` seconds (default **180**),
+which must outlast the server's own startup-readiness budget
+(`RUSTFS_STARTUP_READINESS_MAX_WAIT_SECS`, default 120s) — a shorter poll on a
+slow shared runner mis-reports a slow cold start as a failure. In local mode the
+rig also fails fast if the server process exits before becoming healthy instead
+of polling out the full budget.
+
 ## Scope note
 
 The gate logic is unit-validated across pass/warn/fail/exempt outcomes; the