jwriter performance

[OMosk]

Hello! First of all, thank you for your work on the project. I was curious about go-jsonstream jwriter and I've run your benchmarks, but the results seem to be unexpected.

I was running benchmarks from latest commit (at the moment of Feb 23 2025) in v3 branch, id = 3700efa9d313e4351234afccc117f0e78fd2201d

go version go1.23.4 linux/amd64

goos: linux
goarch: amd64
pkg: github.com/launchdarkly/go-jsonstream/v3/jwriter
cpu: Intel(R) Core(TM) i7-10870H CPU @ 2.20GHz
BenchmarkJSONMarshalComparatives/Null-16                	11927364	        93.33 ns/op
BenchmarkJSONMarshalComparatives/Boolean-16             	 7932656	       152.1 ns/op
BenchmarkJSONMarshalComparatives/NumberInt-16           	 6426847	       173.8 ns/op
BenchmarkJSONMarshalComparatives/NumberFloat-16         	 4020000	       295.5 ns/op
BenchmarkJSONMarshalComparatives/String-16              	 4988412	       241.7 ns/op
BenchmarkJSONMarshalComparatives/ArrayOfBools-16        	  489166	      2291 ns/op
BenchmarkJSONMarshalComparatives/ArrayOfStrings-16      	  151531	      8041 ns/op
BenchmarkJSONMarshalComparatives/Object-16              	 2539591	       463.7 ns/op
BenchmarkWriteNull-16                                   	13551864	       109.7 ns/op
BenchmarkWriteBoolean-16                                	10876502	       120.3 ns/op
BenchmarkWriteNumberInt-16                              	 7821781	       142.8 ns/op
BenchmarkWriteNumberFloat-16                            	 4892714	       242.4 ns/op
BenchmarkWriteString-16                                 	10094972	       121.9 ns/op
BenchmarkWriteArrayOfBools-16                           	  334783	      4070 ns/op
BenchmarkWriteArrayOfStrings-16                         	   73909	     16104 ns/op
BenchmarkWriteObject-16                                 	 3384001	       351.4 ns/op
BenchmarkWriteObjectToNoOpWriterNoAllocs-16             	61757113	        19.59 ns/op
BenchmarkStreamingWriterArrayOfStrings-16               	   79692	     18884 ns/op

I expected streaming writer to vastly outperform encoding/json package, but it is not the case, in some cases it is significantly slower even than encoding/json.

Is it expected?