opt: stabilize end-to-end benchmarks

Previously, end-to-end benchmark results were erratic. Automatic
collection of index recommendations caused the first run of a query
to perform much more poorly than later runs. Also, slow queries which
were benchmarked with few iterations could have dramatically different
results if they were run with over 6 iterations and generic query plans
were selected. Results could look like:

```
BenchmarkEndToEnd/...   6  237211938 ns/op  122903316 B/op  811648 allocs/op
BenchmarkEndToEnd/...   6  191910097 ns/op  100521821 B/op  597137 allocs/op
BenchmarkEndToEnd/...   7  158093143 ns/op  86201213 B/op   512044 allocs/op
BenchmarkEndToEnd/...   8  126410552 ns/op  76577004 B/op   455492 allocs/op
BenchmarkEndToEnd/...  12   83743316 ns/op  52329982 B/op   313305 allocs/op
```

These benchmarks are specifically meant to measure the performance of
query optimization and building logical plans, so generic query plans
and automatic collection of index recommendations have been disabled.
This stabilizes the benchmarks. Now results look more like:

```
BenchmarkEndToEnd/...  6  190383688 ns/op  100462772 B/op  598004 allocs/op
BenchmarkEndToEnd/...  6  190895507 ns/op  100107917 B/op  596267 allocs/op
BenchmarkEndToEnd/...  6  187822812 ns/op  100012398 B/op  595454 allocs/op
BenchmarkEndToEnd/...  6  191119153 ns/op  102148360 B/op  606517 allocs/op
BenchmarkEndToEnd/...  6  202094812 ns/op  100020434 B/op  595465 allocs/op
```

Release note: None

Author: mgartner

pkg/sql/opt/bench/bench_test.go

@@ -1141,6 +1141,8 @@ func BenchmarkEndToEnd(b *testing.B) {
 	sr.Exec(b, `SET CLUSTER SETTING sql.stats.automatic_collection.enabled = false`)
 	sr.Exec(b, `SET CLUSTER SETTING sql.stats.flush.enabled = false`)
 	sr.Exec(b, `SET CLUSTER SETTING sql.metrics.statement_details.enabled = false`)
+	sr.Exec(b, `SET CLUSTER SETTING sql.metrics.statement_details.index_recommendation_collection.enabled = false`)
+	sr.Exec(b, `SET plan_cache_mode = force_custom_plan`)
 	sr.Exec(b, `CREATE DATABASE bench`)
 	for _, schema := range schemas {
 		sr.Exec(b, schema)