Add json transcoding benchmarks

[Ten0]

Resolves #129

Also allows serializing unit as null to enable transcoding (non-hinted deserializers will typically yield unit when encountering null).

Current benchmark results:
(TL;DR going through serde_json seems very slow compared to arrow-json's tailored JSON parser)

json_to_arrow(10000)/arrow-json
                        time:   [3.4072 ms 3.4137 ms 3.4224 ms]

json_to_arrow(10000)/serde_json_transcode_arrow-json
                        time:   [32.399 ms 32.532 ms 32.774 ms]

json_to_arrow(10000)/serde_json_transcode_serde_arrow
                        time:   [30.906 ms 30.947 ms 30.988 ms]

[chmp]

Hey @Ten0,

Thanks a lot for the PR. The performance is unfortunate. But good to know. Overall the changes look good to me. I will add a couple of comments to the code. Some meta comments:

• Not sure how other repos are handling attribution. What would you prefer? My suggestion would be to add "and contributors" to the copyright notice in the license and then maybe add a contributors section in the readme or a separate Authors.md file.
• As you changed the error messages: one of the tests fail (as the message was not that good to begin with, the change itself fine :))
• My guess would be that the additional benchmark will confuse the readme update. I will have a look myself and maybe filter out the JSON benchmarks

[chmp]

* Not sure how other repos are handling attribution. What would you prefer? My suggestion would be to add "and contributors" to the copyright notice in the license and then maybe add a contributors section in the readme or a separate Authors.md file.

• The readme include an "and contributors"

* As you changed the error messages: one of the tests fail (as the message was not that good to begin with, the change itself fine :))

• Fixed

* My guess would be that the additional benchmark will confuse the readme update. I will have a look myself and maybe filter out the JSON benchmarks

• Fixed

[chmp]

@Ten0 in case you're interested. I also added simd_json and rewrote all benchmarks to use the JSON line repr. These are the numbers, I get:

label	time [ms]	arrow_json	serde_arrow (se	arrow_json (ser	serde_arrow (si	serde_arrow (se
arrow_json	8.25	1.00	0.12	0.10	0.06	0.04
serde_arrow (serde_json,transcode)	70.56	8.56	1.00	0.89	0.55	0.33
arrow_json (serde_json,transcode)	79.63	9.66	1.13	1.00	0.62	0.37
serde_arrow (simd_json,transcode)	127.99	15.52	1.81	1.61	1.00	0.60
serde_arrow (serde_json,value)	214.16	25.97	3.03	2.69	1.67	1.00

[chmp]

Thanks for PR!

[Ten0]

Thanks for closing it out! 😊