Fastpath for encoding unsorted dict to JSON

[aivarsk]

JSON encoder can produce output sorted by key (json.dumps(sort_keys=True)) The default value for sort_keys is False. The code always creates a list of dictionary items, sorts them if needed, and then iterates through them.
However, if sorting is not requested we can use PyDict_Next to avoid allocating a list of tuples.

pyperformance on my machine before changes:

### json_dumps ###
Mean +- std dev: 10.6 ms +- 0.1 ms

pyperformance after changes:

### json_dumps ###
Mean +- std dev: 9.00 ms +- 0.08 ms

[aivarsk]

Patched pyperformance to use json.dumps(sort_keys=True)

before changes

### json_dumps ###
Mean +- std dev: 14.0 ms +- 0.2 ms

after changes

### json_dumps ###
Mean +- std dev: 13.6 ms +- 0.2 ms

Probably because the new code does not create an iterator from the list and PyList_GET_ITEM is a bit faster than PyIter_Next (extra incref/decref)?

[serhiy-storchaka]

It caused a regression for OrderedDict. See #100272.