Replaced void pointers with Types in JSON Datetime Conversions

[WillAyd]

This also incorporates the clean up from #30271

Benchmarks show the following benefit, though most likely just noise:

.       before           after         ratio
     [04fce81f]       [f8ed75d0]
     <stash~6^2>       <json-dates>
-         336±9ms          302±4ms     0.90  io.json.ToJSON.time_to_json('columns', 'df_int_floats')

SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY.
PERFORMANCE INCREASED.

One follow up to this can drastically speed up datetime index handling - right now these are done in the Python space but can reuse these same functions

[WillAyd]

The diff is a little difficult to read here, but this got rid of the ToJSON functions and instead replaces them with explicit ToIso or ToEpoch functions with strong types

[jbrockmendel]

how would we get here? if len == 0?

[WillAyd]

This would happen if malloc fails

https://stackoverflow.com/a/12434865/621736

[jbrockmendel]

So failing to address it would mean getting something like a segfault? the PyErr_NoMemory just below raises back in py-space?

[WillAyd]

I think it would segfault if you tried to access the memory if malloc failed to allocate it. PyErr_NoMemory would just set the global error indicator so when the extension exits that should raise back in the Python space (assuming nothing else clears it out)

FWIW the error handling here is copy / paste from here:

pandas/pandas/_libs/src/ujson/python/objToJSON.c

Line 422 in 98cb432

PyErr_NoMemory();

There are similar checks elsewhere throughout the module (not always consistent) so maybe worth unifying in a follow up

[jbrockmendel]

PyUnicode_AsUTF8AndSize came up in the thread about unicode surrogates. do we need to worry about those here?

[jbrockmendel]

@WillAyd i think this is the same function where passing a surrogate unicode character can cause a segfault. can that be ruled out here?

[jbrockmendel]

With usual caveat about my C-fu being weak, LGTM. cc @jreback

[jbrockmendel]

@jreback gentle ping, just realized this is a blocker for #28595.

[jreback]

thanks @WillAyd looks more clear