sql/covering: make OverlapCoveringMerge benchmark deterministic

Previously we seeded RNG with time and benchmark results varied widely.
In order to make sure this doesn't make noise when comparing different
versions make the seed a constant.

Fixes: #88919

Release note: None

pkg/sql/covering/BUILD.bazel

@@ -20,7 +20,6 @@ go_test(
     embed = [":covering"],
     deps = [
         "//pkg/util/leaktest",
-        "//pkg/util/timeutil",
         "@com_github_stretchr_testify//require",
     ],
 )

pkg/sql/covering/bench_test.go

@@ -16,7 +16,6 @@ import (
 	"math/rand"
 	"testing"
 
-	"github.com/cockroachdb/cockroach/pkg/util/timeutil"
 	"github.com/stretchr/testify/require"
 )
 
@@ -25,7 +24,7 @@ func BenchmarkOverlapCoveringMerge(b *testing.B) {
 		name   string
 		inputs []Covering
 	}
-	rand.Seed(timeutil.Now().Unix())
+	rand.Seed(0)
 
 	for _, numLayers := range []int{
 		1,      // single backup