[ZeroIntensity]

The idea here is that when a list is created with PyList_New(n) where n is greater than zero, the list cannot be shared across threads, because it's unusable as an object (due to the NULL pointers that it would try to access upon use), and therefore we can skip locking.

This was generated by AI without my review; I'm putting this up as a proof of concept that we can use for benchmarking and such.

• Issue: PyList_SetItem is ~2.7x slower per item on the free-threaded build #158660

[ZeroIntensity]

Benchmarks, using Nathan's fill_list script from the issue:

• This PR, free-threaded: 3.4 ns/item
• Main, free-threaded: 4.6 ns/item
• This PR, GIL-icious: 2.5 ns/item
• Main, GIL-icious: 2.5 ns/item

So, with this fix, PyList_SetItem is now about 35% slower than the default build, in contrast to about 84% slower without it. I'll look into ways to optimize this further tomorrow.