[Core] Experimental support for no-GIL build and JIT in Python 3.13

[Superskyyy]

Description

With Python 3.13 officially out and two "killer features" that aim to boost the experience of ML and data science users, we could attempt to see if the features can be safely enabled on Ray, thus leading to further adoption. Major libs have already been pushing towards this goal.

Use case

free-threading in Python 3.13 allows concurrent execution without the constraints of a GIL natively in Python, which could trigger some new usages in Ray. On JIT side just to make sure every test still passes would be enough and see if there are any performance gains.

[aslonnie]

cc @anyscalesam

[Superskyyy]

Note that both features are disabled by default in a normal CPython build; we need to enable them explicitly to try them out.

[cclauss]

Perhaps there should be two separate focuses.

1. Compatibility and wheels for out-of-the-box Py3.13.
2. Compatibility and wheels for the experimental free-threaded Py3.13-t.

My bet would be that 1. could be delivered quickly while 2. will take more time and effort.

https://docs.python.org/3.13/whatsnew/3.13.html#whatsnew313-free-threaded-cpython

[pcmoritz]

For people interested in this, there is a prototype in #47984 you can use to build Ray yourself on Python 3.13 (with pip install -e .) and try it out :)

[caoergou]

Looking forward to Ray with Python 3.13 without GIL!