Improve ISO Date Performance for JSON

[WillAyd]

benchmarks below

       before           after         ratio
     [9c6771c5]       [5c0f5682]
     <master>         <json-index-dates>
-        231±60ms          188±3ms     0.82  io.json.ToJSON.time_iso_format('split', 'df_td_int_ts')
-        496±20ms         218±40ms     0.44  io.json.ToJSON.time_iso_format('index', 'df_int_float_str')
-        486±20ms          207±2ms     0.42  io.json.ToJSON.time_iso_format('columns', 'df_int_float_str')
-         499±9ms          210±1ms     0.42  io.json.ToJSON.time_iso_format('index', 'df_date_idx')
-        503±20ms          206±2ms     0.41  io.json.ToJSON.time_iso_format('columns', 'df_int_floats')
-        515±20ms          210±1ms     0.41  io.json.ToJSON.time_iso_format('index', 'df')
-        524±80ms        209±0.5ms     0.40  io.json.ToJSON.time_iso_format('index', 'df_int_floats')
-        528±10ms          206±2ms     0.39  io.json.ToJSON.time_iso_format('columns', 'df_td_int_ts')
diff --git a/asv_bench/asv.conf.json b/asv_bench/asv.conf.json
index c04bbf53a..897de2f85 100644
--- a/asv_bench/asv.conf.json
+++ b/asv_bench/asv.conf.json
-        546±80ms        209±0.7ms     0.38  io.json.ToJSON.time_iso_format('index', 'df_td_int_ts')
-        568±60ms          208±3ms     0.37  io.json.ToJSON.time_iso_format('columns', 'df_date_idx')
-       598±100ms          206±2ms     0.35  io.json.ToJSON.time_iso_format('columns', 'df')

SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY.
PERFORMANCE INCREASED.

Note that this mostly improves on DTI which on 0.25.3 can't even be written as ISO format, so I didn't add a whatsnew. Timedelta is the big bottleneck remaining

[jbrockmendel]

CI notwithstanding, this looks like a nice cleanup+speedup

[pep8speaks]

Hello @WillAyd! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2020-01-02 00:00:36 UTC

[WillAyd]

In another follow up I think I'll move the conversion routines to another file and keep only the core JSON serialization functionality within this ones; makes it a little easier to grok the difference between functions used as callbacks and those used to convert values into various formats

[jbrockmendel]

LGTM

[jreback]

do you need a Py_DECREF anywhere here?

[WillAyd]

Don't think so - here's a snippet in NumPy where there is no DECREF

https://github.com/numpy/numpy/blob/5ce770ae3de63861c768229573397cadd052f712/numpy/core/src/multiarray/scalarapi.c#L212

And testing locally did segfault when trying

[jreback]

thanks, might be worth a general whatsnew note for perf for json changes (e.g. list issues that impacted perf that you have done recently)

[WillAyd]

Yea no problem; will address in follow up