Use orjson for our json renderer

[dstufft]

This switches our pyramid renderer from using the stdlib json module, to using orjson instead.

I ran benchmarks testing the speed of dumps on the JSON version of /simple/ from live PyPI, and I got these results:

--------------------------------------------------------------------------- benchmark 'pypi.json serialization': 2 tests ---------------------------------------------------------------------------
Name (time in ms)                     Min                 Max                Mean            StdDev              Median               IQR            Outliers      OPS            Rounds  Iterations
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
test_dumps[pypi.json-orjson]      19.4031 (1.0)       20.7105 (1.0)       19.8266 (1.0)      0.5764 (1.0)       19.4697 (1.0)      0.8415 (1.0)           1;0  50.4373 (1.0)           5         100
test_dumps[pypi.json-json]       120.4422 (6.21)     122.5178 (5.92)     121.7277 (6.14)     0.8201 (1.42)     121.8010 (6.26)     1.1179 (1.33)          1;0   8.2151 (0.16)          5          10
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

This should make us something like 6x faster on serializing our json endpoints according to the above benchmarks.

The one gotcha with using orjson that makes it not a drop in replacement for json is that instead of returning a str it returns bytes. This isn't a big deal for us as Pyramid's render system supports both, and will just encode str results anyways into bytes.