remove continue-on-error from free-threaded CI job

[ngoldbaum]

Now that 3.13.0rc2 is out, the tests shouldn't be crashing anymore.

In principle it's also possible internal uses of GILOnceCell could also cause crashes. In practice I don't ever see any crashes caused by GILOnceCell in my local testing.

I do need help with one thing, there is currently a failing test introduced by #4453, which added a new diagnostic warning to compiler errors if the abi3 feature is enabled. But since the free-threaded build (like pypy) simply warns that abi3 isn't supported and ignores it, this leads to a test failure due to this new line in the compiler error:

$ diff -u expected actual
--- expected	2024-09-11 14:26:36
+++ actual	2024-09-11 14:26:47
@@ -6,6 +6,7 @@
   |
   = help: the trait `PyClassBaseType` is not implemented for `PyBool`
   = note: `PyBool` must have `#[pyclass(subclass)]` to be eligible for subclassing
+  = note: with the `abi3` feature enabled, PyO3 does not support subclassing native types
   = help: the following other types implement trait `PyClassBaseType`:
             PyAny
             PyArithmeticError
@@ -30,6 +31,7 @@
   |
   = help: the trait `PyClassBaseType` is not implemented for `PyBool`
   = note: `PyBool` must have `#[pyclass(subclass)]` to be eligible for subclassing
+  = note: with the `abi3` feature enabled, PyO3 does not support subclassing native types
   = help: the following other types implement trait `PyClassBaseType`:
             PyAny
             PyArithmeticError

I'm not sure how to account for that in the compiler error tests.

[davidhewitt]

Nice! I pushed a commit to fix the diagnostic (realised it was easier to push than to describe).

[davidhewitt]

Looks like there's a (new?) failure in test_gc 🤔

[ngoldbaum]

I can reproduce the failure on an x86 linux machine. I wonder why it only happens on x86 linux so far, I'll try to understand more.

[ngoldbaum]

I ignored the assert that was failing, I've convinced myself it is not a valid thing to assert on the free-threaded build and left a comment to that effect.

I don't see any other crashes on linux besides python/cpython#124375 (I ran into and reported it working on this PR), which I think is unrelated to PyO3.

[davidhewitt]

Makes sense, thanks. Think there's just one thing which might have changed too far...