json: fixes incorrect handling of error tag

[codefromthecrypt]

Before, we didn't document that an existing tag with the same name as the errorTag parameter of the JSON writer would be retained. This documents it. It also fixes a bug where that key is not literally "error", for example, adding a tag named "exception".

I backfilled tests and adjusted benchmarks to make sure the result isn't worse than before.

Before

Benchmark                                                          Mode     Cnt    Score   Error   Units
ZipkinV2JsonWriterBenchmarks.sizeInBytes_bigClientSpan:gc.alloc.rate.norm  sample      15   0.042 ± 0.007    B/op
ZipkinV2JsonWriterBenchmarks.sizeInBytes_bigClientSpan:p0.99               sample           0.334           us/op
ZipkinV2JsonWriterBenchmarks.write_bigClientSpan:gc.alloc.rate.norm        sample      15  24.105 ± 0.008    B/op
ZipkinV2JsonWriterBenchmarks.write_bigClientSpan:p0.99                     sample           0.791           us/op

After

Benchmark                                                                    Mode     Cnt    Score   Error   Units
ZipkinV2JsonWriterBenchmarks.sizeInBytes_bigClientSpan:gc.alloc.rate.norm  sample      15    0.041 ± 0.006    B/op
ZipkinV2JsonWriterBenchmarks.sizeInBytes_bigClientSpan:p0.99               sample            0.334           us/op
ZipkinV2JsonWriterBenchmarks.write_bigClientSpan:gc.alloc.rate.norm        sample      15   24.101 ± 0.017    B/op
ZipkinV2JsonWriterBenchmarks.write_bigClientSpan:p0.99                     sample            0.750           us/op

[codefromthecrypt]

merging to unblock