Refactor Batch Processor benchmark to really test OnEnd (#5600)

This benchmark was not only testing the batch processor, since it was
starting a span, and ending it.
So the entire process of recording span was being accounted.

This changes the benchmark to only account for the batch processor being
tested, not the entire stack.

Before:
```
pkg: go.opentelemetry.io/otel/sdk/trace
BenchmarkSpanProcessor-10                         137320              8696 ns/op           11184 B/op        155 allocs/op
BenchmarkSpanProcessorVerboseLogging-10           135483              8881 ns/op           11184 B/op        155 allocs/op
PASS
ok      go.opentelemetry.io/otel/sdk/trace      3.984s
```

After:
```
pkg: go.opentelemetry.io/otel/sdk/trace
BenchmarkSpanProcessorOnEnd/batch:_10,_spans:_10-10              6055572               173.2 ns/op             0 B/op          0 allocs/op
BenchmarkSpanProcessorOnEnd/batch:_10,_spans:_100-10              684236              1733 ns/op               0 B/op          0 allocs/op
BenchmarkSpanProcessorOnEnd/batch:_100,_spans:_10-10             6930391               173.8 ns/op             0 B/op          0 allocs/op
BenchmarkSpanProcessorOnEnd/batch:_100,_spans:_100-10             677128              1731 ns/op               0 B/op          0 allocs/op
BenchmarkSpanProcessorVerboseLogging-10                           128823              9318 ns/op           11184 B/op        155 allocs/op
PASS
ok      go.opentelemetry.io/otel/sdk/trace      6.763s
```

I haven't touched the verbose logging benchmark, as I suppose a
benchmark with verbose logging could actually benefit from the entire
stack.
I also feel one benchmark testing the entire stack up to there could be
nice.

sdk/trace/batch_span_processor_test.go

@@ -627,23 +627,32 @@ func TestBatchSpanProcessorConcurrentSafe(t *testing.T) {
 	wg.Wait()
 }
 
-func BenchmarkSpanProcessor(b *testing.B) {
-	tp := sdktrace.NewTracerProvider(
-		sdktrace.WithBatcher(
-			tracetest.NewNoopExporter(),
-			sdktrace.WithMaxExportBatchSize(10),
-		))
-	tracer := tp.Tracer("bench")
-	ctx := context.Background()
-
-	b.ResetTimer()
-	b.ReportAllocs()
-
-	for i := 0; i < b.N; i++ {
-		for j := 0; j < 10; j++ {
-			_, span := tracer.Start(ctx, "bench")
-			span.End()
-		}
+func BenchmarkSpanProcessorOnEnd(b *testing.B) {
+	for _, bb := range []struct {
+		batchSize  int
+		spansCount int
+	}{
+		{batchSize: 10, spansCount: 10},
+		{batchSize: 10, spansCount: 100},
+		{batchSize: 100, spansCount: 10},
+		{batchSize: 100, spansCount: 100},
+	} {
+		b.Run(fmt.Sprintf("batch: %d, spans: %d", bb.batchSize, bb.spansCount), func(b *testing.B) {
+			bsp := sdktrace.NewBatchSpanProcessor(
+				tracetest.NewNoopExporter(),
+				sdktrace.WithMaxExportBatchSize(bb.batchSize),
+			)
+			snap := tracetest.SpanStub{}.Snapshot()
+
+			b.ResetTimer()
+			b.ReportAllocs()
+			for i := 0; i < b.N; i++ {
+				// Ensure the export happens for every run
+				for j := 0; j < bb.spansCount; j++ {
+					bsp.OnEnd(snap)
+				}
+			}
+		})
 	}
 }