roachtest/perturbation: lower ratioOfMax to 0.3 for restart

The restart perturbation runs the workload at 50% of cluster peak,
calibrated for steady state. During recovery the target node must
absorb raft catch-up, rebalancing snapshots, and re-acquired lease
traffic on top of its share of the ongoing workload, which can push
it past disk capacity into a Pebble write stall. The stall blocks
every leaseholder that lives on the recovering store and produces a
recovery-phase p5 throughput dip that trips the 1.25x impact gate.

Drop ratioOfMax to 0.3 for restart only so the cluster runs with ~40%
IO headroom; recovery work then fits within sustainable disk capacity.

Fixes: #170849
Release note: None

Co-Authored-By: roachdev-claude <roachdev-claude-bot@cockroachlabs.com>

Author: miraradeva

pkg/cmd/roachtest/tests/perturbation/restart_node.go

@@ -27,6 +27,13 @@ func (r restart) setup() variations {
 	r.cleanRestart = true
 	v := setup(r, defaultThresholds())
 
+	// Run with extra IO headroom relative to the cluster-wide default. When the
+	// target node returns it must absorb raft catch-up, rebalancing snapshots,
+	// and re-acquired lease traffic concurrently; at the default 0.5 the
+	// recovered store sits at the edge of its sustainable write rate and the
+	// flush/compaction pipeline can fall behind into a Pebble write stall.
+	v.ratioOfMax = 0.3
+
 	// TODO(baptist): Remove this setting once #120073 is fixed.
 	v.clusterSettings["kv.lease.reject_on_leader_unknown.enabled"] = "true"