gh-132312: Significantly reduce thread state lookups in the interpreter

[ZeroIntensity]

This does a lot of refactoring to avoid repeated _PyThreadState_GET calls in hot code paths (primarily the eval loop + common object operations).

Most of the changes here are mechanical, but I did have to make some more convoluted changes to make this robust (such as the addition of _Py_DECREF and such).

• Issue: Add a PyThreadState * parameter (almost) everywhere #132312

[picnixz]

Do you have some numbers on how much we gain with this dance?

[ZeroIntensity]

Nothing conclusive yet, no.

[ZeroIntensity]

I'm having a lot of trouble getting consistent benchmarks, unfortunately. My system seems to be too jittery to get a reliable result. I also don't know why the JIT CI is failing on Windows.

@Fidget-Spinner -- do you have advice for either of these?

[Fidget-Spinner]

@ZeroIntensity JIT on Windows requires whatever is used in bytecodes.c to be an exported symbol. In this case it's complaining memcpy is not available. Are you using that? If so, you need to add the relevant header to template.c

[ZeroIntensity]

I didn't add anything that uses memcpy in this PR, but shouldn't functions from the C stdlib be considered exported anyway?

[Fidget-Spinner]

It could be indirectly used as well through an inlined function. In any case, unless string.h is included by Python.h or the other pycores by default it wont be included https://github.com/python/cpython/blob/main/Tools/jit/template.c