Fix memory leak with ujson module

[WillAyd]

• closes BUG: Reference leak in pandas/_libs/src/ujson/python/objToJSON.c #49222 (Replace xxxx with the GitHub issue number)

This is mostly vendored from ujson upstream

https://github.com/ultrajson/ultrajson/blob/main/python/ujson.c

Using a static PyObject the way we did before is strongly discouraged. This is more verbose but should be more correct, although interestingly pushes us back to the legacy module initialization instead of PEP 489 multi phase initialization

[WillAyd]

@JelleZijlstra if you have the ability to confirm this patch fixes your issue would be appreciated

[akx]

As an aside, has Pandas considered https://github.com/ijl/orjson (a Rust-based JSON (de)serializer)?

[mroeschke]

Thanks @WillAyd. Do you think it's worth adding and ASV peakmem_* benchmarks or are we sufficiently covered?

[WillAyd]

Its a pretty small memory leak. I don't think it would even register on a peakmem ASV. Let's see if it helps @JelleZijlstra and can go from there

[JelleZijlstra]

Thanks, I can confirm this removes the leak. I left a few comments on the code.

@mroeschke this memleak only appears when pandas is unimported from a running process. It's unlikely to affect many current use cases.

[mroeschke]

Looks good @WillAyd. Would be good to have a whatsnew note for this then LGTM

[mroeschke]

Thanks @WillAyd

[rhshadrach]

This patch may have induced a potential regression. Please check the links below. If any ASVs are parameterized, the combinations of parameters that a regression has been detected appear as subbullets. This is a partially automated message.

• https://asv-runner.github.io/asv-collection/pandas/#io.json.ToJSON.time_to_json
  • orient='split'; frame='df_int_float_str'
  • orient='split'; frame='df_int_floats'
  • orient='split'; frame='df_td_int_ts'
  • orient='values'; frame='df_int_float_str'
  • orient='values'; frame='df_int_floats'
  • orient='values'; frame='df_td_int_ts'
• https://asv-runner.github.io/asv-collection/pandas/#io.json.ToJSONWide.time_to_json
  • orient='split'; frame='df_int_floats'
  • orient='split'; frame='df_td_int_ts'
  • orient='values'; frame='df_int_float_str'
  • orient='values'; frame='df_int_floats'
  • orient='values'; frame='df_td_int_ts'